Cross-domain modeling method and system for unified semantic objects

Abstract

The application discloses a cross-domain modeling method and system of unified semantic objects, and belongs to the fields of artificial intelligence and computer-aided design. Multi-source semantic input generates additional immutable semantic evidence records, and each of the records contains positioning information and candidate hypotheses; the records are fused after being anchored to entity, relationship or constraint type unified semantic objects, and non-converged records are left as candidate hypotheses; a unified semantic projection range is formed, and completeness is determined for each target; a domain model is generated by a projection operator for records meeting generation conditions, and a projection relationship record of semantic elements and model elements is established; semantic conservation quantities of the domain model and a benchmark are compared, and only consistent records are marked as synchronizable models. When changed, the synchronization qualification is temporarily withdrawn, an atomic model is decomposed by an improvement operator to change and generate a semantic change candidate, and an updated unified semantic object is formed after the change candidate is fused; the projection qualification of the affected is temporarily withdrawn and then incrementally re-projected, and the source model is restored after being closed-loop verified. Cross-domain propagation is performed through unified semantic objects.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and computer-aided design technology, specifically to semantic consistency processing of heterogeneous model data, which can be applied to computer-aided design (CAD / CAE / CAM), systems engineering modeling (MBSE), software architecture modeling, digital twins, 3D scene construction and human-computer interactive modeling and other scenarios.

[0002] In particular, this invention relates to a cross-domain modeling method, system, and computer-readable storage medium that uses a dynamic semantic state field and a unified semantic object as a semantic foundation. This method involves reconstructing a unified semantic object from semantic input data from one or more information sources, such as text, speech, images, video, sensor data, program code, existing model files, external AI model output, and user interaction behavior, through semantic evidence recording and evidence fusion. The unified semantic object is then consistently projected onto a target domain using a unified semantic projection range as the processing unit, undergoes bidirectional synchronization after semantic conservation verification, and continuously evolves. The system of this invention supports consistent projection onto at least two registered target domains, while the current modeling task can be performed on only one or more target projections. After the formal domain models of at least two different target domains obtain synchronization eligibility, a cross-domain synchronization relationship is formed between them through the unified semantic object. Background Technology

[0003] The digital description of complex engineering objects and virtual environments typically spans multiple modeling paradigms: in systems engineering, SysML, UML, Modelica, etc., are used to describe structure, function, behavior, and constraints; in industrial design, CAD / CAE is used to describe geometry, topology, and manufacturing constraints; in software, component diagrams and node diagrams are used to describe architecture and interfaces; in the 3D content field, scene diagrams and entity-component paradigms are used to describe space, materials, and interactions; in recent years, large language models and multimodal models have been able to generate graphics, code, or scenes from text and images.

[0004] However, the existing technology has the following drawbacks.

[0005] Defect 1: Lack of a unified semantic foundation across domains; model synchronization relies on direct point-to-point mapping. In architectures requiring independent bidirectional mappings for any two domains, the number of mapping rules increases exponentially with the number of target domains. n D according to O ( n D 2The growth and maintenance costs are high; moreover, such mappings are mostly syntactic-level format correspondences, making it difficult to verify deep semantic equivalence, and modifications in one place are difficult to propagate automatically and correctly across domains. Model transformation and data exchange schemes based on predefined metamodels rely on the pre-binding of source and target metamodels, and usually lack a universal semantic reference system independent of any modeling language.

[0006] Defect 2: Some shared repository architectures lack gating discipline. In common implementations of architectures that coordinate multiple components using a central shared data structure (such as blackboard architectures and global state repositories), writes are usually not preconditioned by whether the semantics are sufficiently determined, whether the information is sufficient to generate the target product, or whether the generated product faithfully adheres to the shared semantics: the component's conclusion is immediately visible to downstream components once written. Such implementations typically focus on decoupling direct communication between components and providing a unified data access interface, but struggle to simultaneously provide an immutable evidence layer that decouples generation and acceptance, a multiple hypothesis suspension mechanism, completeness judgment relative to the generation target, and consistent gating of the generated product's adherence to the shared semantics. Even if some implementations have access control, version management, or data format validation, their validation targets are usually write permissions, concurrency timing, and data structure legality, rather than the three types of semantic criteria mentioned above. Therefore, erroneous conclusions may still propagate downstream.

[0007] Defect 3: Multi-view synchronization technologies employ direct inter-view linkage paths. Some multi-view synchronization solutions within the same tool typically establish direct coordinate-level or element-level mappings between the datasets of two views. Modifications to one view are directly applied to the other view via pre-bound rules. Such point-to-point paths struggle to accommodate uncertain information from non-geometric modalities such as text, images, and audio, and are difficult to express deep semantics such as object identity and interface signatures. They lack evidence backtracking and arbitration mechanisms when synchronization fails, and the maintenance cost of mapping rules increases dramatically when extended to heterogeneous domains.

[0008] Defect 4: AI-generated results lack a credible engineering closed loop. Some "input-direct-model generation" technologies are unidirectional and one-off: the reasons why model elements are generated cannot be traced, the confidence level is unclear, there is no arbitration mechanism when information from different sources conflicts, the generated results are not subject to engineering constraints and verification, and the illusions or errors of the AI ​​model can directly contaminate the model and cannot be corrected without loss. Schemes that directly write the output of large models into knowledge graphs or model libraries also lack an immutable evidence layer and acceptance gating.

[0009] Defect 5: Inability to determine when the description is "complete enough". Existing solutions typically lack a definition of computable completeness relative to the generated target: the system neither knows whether the current information is sufficient to generate a certain type of target model, nor what information is missing, or at what cost to obtain it from which source, leading to a large amount of invalid generation and repeated rework.

[0010] Defect 6: Digital twin-type single-fact-source schemes lack a conservation verification loop. In common implementations of such schemes, although a unified central state is maintained for multiple parties to read, the consistency between the generated product and the central semantics usually lacks computable criteria and pre-gating, and there is no verification after generation, and no repair or degradation path after failure.

[0011] Defect 7: Lack of semantic-level baselines and idempotent control for concurrent changes. Complex modeling tasks are often carried out in parallel by multiple engineers, multiple tools, and even automated processes. The projections of the same object in different domains may be edited simultaneously. Existing solutions generally adopt write-after-over, pessimistic locking, or syntax-level three-way merging to address this: write-after-over directly erases previously accepted results based on outdated baselines, resulting in lost modifications; pessimistic locking sacrifices parallelism for safety, which contradicts the actual working methods of multi-domain collaboration; syntax-level three-way merging (such as text line difference merging, model element difference merging) can only determine whether the same position has been modified simultaneously, but cannot determine two modifications in different domains that are syntactically independent but semantically mutually exclusive. For example, one party adds a bus in the hardware architecture, and the other party lowers the power budget in the requirements document. Both are legal on their own, but after merging, they both violate the power constraint, while syntax merging will not report any conflict. Furthermore, existing solutions typically lack a baseline semantic version for their change logs. Consequently, the system cannot determine whether a change is based on an outdated semantic state, and therefore cannot reinterpret the change when the semantics have changed. When change messages are retransmitted over the network or repeatedly sent back by tools, they also lack idempotency indicators, potentially leading to the same change being accepted repeatedly. Moreover, when multiple change events affect the same semantic unit, existing solutions trigger downstream synchronization separately for each event, resulting in redundant projections and inconsistencies in intermediate states. Downstream domains may briefly see contradictory model versions. The root of these problems lies in the fact that concurrency control is placed at the file or model element level, rather than at a unified semantic level. The system lacks both a semantic-level baseline version and semantic-level conflict criteria and merging units.

[0012] In summary, there is an urgent need in this field for a cross-domain modeling technology that is centered on an independent semantic foundation, uses immutable evidence as input discipline, employs completeness and semantic conservation as dual gating, uses evidence-based backflow as a synchronization mechanism, and achieves concurrent safety with semantic-level benchmark versions and merging units. This technology would enable heterogeneous multi-source inputs to be uniformly restored into a computable, traceable, and evolvable semantic representation, and projected consistently to multiple target domains while being securely synchronized bidirectionally. Summary of the Invention

[0013] Terminology Definition

[0014] Information Sources and Input Data: Information sources refer to the sources from which the system provides input data, including but not limited to user input (text, voice, sketches, interactive operations), files (requirement documents, drawings, existing domain model files, program code), sensors and devices, and external artificial intelligence models; Input data refers to the original information that has been normalized by the input adapter of the access and interaction layer, carrying source identifiers and timestamps, and is divided into two categories according to its function: semantic input data and control configuration input.

[0015] Semantic evidence records: Structured data records generated by the evidence generation module for semantic input data, including at least source information, timestamps, and one or more semantic evidence fragments, and may also include extractor identifiers and versions, overall evidence quality scores; candidate semantic hypotheses, the initial evidence support corresponding to each candidate semantic hypothesis, and the corresponding spatiotemporal positioning information are carried separately by the semantic evidence fragments and are not repeatedly carried in the record layer. Semantic evidence records are the carrier for decoupling "generation" and "acceptance": semantic input data used to determine the semantic content of an object in the output of any information source (including external artificial intelligence models) can only participate in the update of a unified semantic object in the form of semantic evidence records, and there is no direct writing path; the control configuration input in the input data does not form semantic evidence records, but is recorded as system operation configuration in the evolution field.

[0016] Append-based immutability: This means that once semantic evidence records and domain change evidence records are written into the evidence field of the dynamic semantic state field, their semantic content will no longer be modified or deleted. Any corrections will be made by adding new records. The immutability mentioned is a constraint at the logical semantic level, and does not restrict optimizations such as compression, archiving, and cold storage at the physical storage level.

[0017] A unified semantic object is the basic unit carrying semantic content in a dynamic semantic state field. It consists of an octet including a confidence function, a typed relational structure, attribute functions that change over time and context, a set of behavioral capabilities, a set of constraints, a set of evidence citations, an evolutionary history, and projection relations. The confidence function provides the current confidence level for any candidate semantic hypothesis of the object, including concept class hypotheses, attribute value hypotheses, and relational hypotheses. The unified semantic object persists independently of any domain model; the domain model is a projection of the unified semantic object, not the object itself.

[0018] The Dynamic Semantic State Field is a global data structure that carries all semantic content of unified semantic modeling and serves as a container for unified semantic objects. It consists of six sub-fields: an entity field for storing entity-type unified semantic objects and their semantic states; a relation field for storing relation-type unified semantic objects and the relationships between them; a constraint field for storing constraint-type unified semantic objects, constraints, and their satisfied states; an evidence field for storing semantic evidence records in an append-only, immutable manner; an evolution field for storing object versions, reasons for changes, and scope of impact; and a hypothesis field for storing undetermined candidate semantic hypotheses and their confidence levels. The Dynamic Semantic State Field itself is a data structure and does not perform calculations; each functional module performs read and write operations around it. Semantic content and business state data are exchanged entirely through the Dynamic Semantic State Field; modules do not directly transmit semantic content and business state data to each other. Modules can directly transmit control signals without semantic content, including event identifiers, task version identifiers, semantic element identifiers, processing requests, and completion notifications. The receiving module reads the corresponding semantic content from the Dynamic Semantic State Field based on the control signals before processing it.

[0019] Semantic Element Types and Materialization: In this invention, the unified semantic object is the only semantic element carrying type, and is divided into three types according to the semantic elements it carries: entity-type unified semantic objects, carrying independently referential objects; relation-type unified semantic objects, carrying typed relationships between two or more unified semantic objects; and constraint-type unified semantic objects, carrying evaluable constraint expressions acting on one or more unified semantic objects and their satisfied states. The terms "relational object" and "constraint object" used in this specification and claims are abbreviations for "relational unified semantic object" and "constraint-type unified semantic object," respectively, and do not constitute other types of data structures independent of the unified semantic object. Materialization refers to the operation of upgrading the relations or constraints held inline within the unified semantic object into independent relational or constraint-type unified semantic objects: Before materialization, relations and constraints exist as inline items in the relation structure and constraint set of the object, respectively, without independent object identifiers, confidence functions, and evolutionary histories; after materialization, the relation or constraint has all the components of the unified semantic object, and can independently carry attributes, constraints, evidence tracing references, and versions, while the corresponding items in the original object's relation structure or constraint set become references to it. The three types are stored in the entity field, relation field, and constraint field of the dynamic semantic state field, respectively, and are equally applicable to all the mechanisms of this invention, such as evidence object anchoring, evidence fusion, convergence criteria, projection relation recording, influence range determination, incremental reprojection, and tombstone state; the evidence field, evolution field, and hypothesis field are common subfields that cross the three types. Furthermore, a single attribute item in the attribute function of a unified semantic object can be used as an independent semantic processing unit, namely, an attribute item semantic unit. It includes at least the attribute item identifier, the unified semantic object identifier to which it belongs, the determined semantic content of the attribute item, the candidate semantic hypothesis group of the attribute item and its confidence level, the current convergence state of the attribute item, the constraint references acting on the attribute item and their satisfaction status, the semantic evidence fragment references supporting the attribute item, the source semantic version of the attribute item, the evolution history of the attribute item, and the projection relationship of the attribute item. The reason why the above fields must be listed in full is that when an attribute item is objectified, it must be migrated or initialized to the unified semantic object to which it is objectified. If the attribute item semantic unit itself does not carry the determined semantic content, convergence state, constraint satisfaction status, and source semantic version, then the source of these contents cannot be determined during objectification, and the migration loses its basis for execution. Accordingly, semantic evidence fragments can be associated with the unified semantic object as a whole, or with its specified attribute items. The convergence of an attribute item depends on whether its own candidate semantic hypothesis group satisfies the convergence criterion, and is independent of the convergence of the unified semantic object to which it belongs. The convergence of the conceptual class hypothesis of the object as a whole does not mean that all its attribute items have converged, and vice versa.Attribute semantic units do not constitute another type of data structure independent of the unified semantic object; their confidence and evolution history are stored under the corresponding sub-items of their respective unified semantic object. When an attribute independently satisfies the convergence criterion, its candidate semantic hypothesis has converged into an entity semantic that can be independently referred to, or it is confirmed by evidence that it corresponds to a component, interface, resource, or other entity with an independent object identity, and has at least one of the following: an independent projection relationship record, an independently recordable evolution history, and an independent set of constraints, it can be objectified into an independent entity-type unified semantic object according to the attribute objectification process described later in this specification (its triggering conditions, migration content, and execution order are explained in the specific implementation section of this specification regarding attribute objectification). "Being able to be independently referred to" is a necessary condition. Since the definition of a unified semantic object is to carry an independently referential object, pure parameter attributes such as "power consumption limit 35W", "color is black" and "weight 3 kg", although they may have independent evolutionary history or independent projection relationship, do not correspond to any independently referential real objects. If they are objectified together, pure value-based objects that do not conform to their definition will appear in the entity field, and the object identity set of the benchmark semantic conservation quantity will be polluted by parameter items. Therefore, such attribute items are retained as attribute items and not objectified. After objectification, the original attribute items are converted into references to the unified semantic object and are not deleted. Except for a re-fusion with migration markers that can be performed during objectification, the migrated semantic evidence fragments will not participate in the fusion of the objectified unified semantic object again to avoid the same observation being counted repeatedly.

[0020] Candidate semantic hypotheses and mutually exclusive candidate semantic hypothesis groups: A candidate semantic hypothesis refers to a possible interpretation of the same semantic information item and its confidence level; candidate semantic hypotheses that cannot be simultaneously true for the same semantic information item to be determined constitute a mutually exclusive candidate semantic hypothesis group; semantic content that has not yet met the convergence condition is retained in the hypothesis field as a candidate semantic hypothesis state and is not written into the determined semantic state. The "determined semantic state" of a unified semantic object as referred to in this specification and claims refers to the semantic content of the object that has met the convergence criterion and can be used for projection generation and cross-domain change propagation, including at least one of semantic state, attribute, relationship, behavioral ability and constraint, which is different from the undetermined candidate semantic hypotheses retained in the hypothesis field; accordingly, "updating the determined semantic state" is not limited to updating the conceptual semantics of the object, but covers the updating of the above-mentioned semantic content as well as the addition and deletion of entity-type, relation-type and constraint-type unified semantic objects.

[0021] Semantic input data and control configuration input: Semantic input data refers to the input data used to determine the semantic content (semantic state, attributes, relationships, behavioral capabilities, constraints) of a unified semantic object; semantic input data from any information source must be recorded as semantic evidence before entering the dynamic semantic state field, and there is no direct write path. Control configuration input refers to the input data used to set the target projection set, various thresholds and weights, access permissions, and result presentation methods. It acts on the system operation configuration and does not participate in fusion as semantic evidence; its changes are recorded in the evolution field. User output can simultaneously include two types: the user's confirmation or modification results of semantic content belong to semantic input data and participate in fusion as highly reliable semantic evidence; the user's settings for the target projection set and thresholds belong to control configuration input.

[0022] Semantic evidence fragment: A semantic unit with independent location information contained in a semantic evidence record; each semantic evidence fragment includes a fragment identifier, location information, candidate semantic hypotheses, and their initial evidence support, and can be associated with different unified semantic objects. When a semantic evidence record describes only a single object, it contains one semantic evidence fragment. Evidence object anchoring, anchoring correction, and replay-based refusion all use semantic evidence fragments as processing units, thereby supporting scenarios where an image, requirement document, or existing model file describes multiple objects simultaneously, as well as scenarios where only some fragments are re-anchored to other objects.

[0023] The unified semantic projection scope is a semantic subset corresponding to the target projection set in this round. It includes one or more entity-type unified semantic objects, and at least one of the following: an entity-type unified semantic object associated with the entity-type unified semantic object, and a constraint-type unified semantic object acting on it. When no relational or constraint unified semantic objects exist, it is composed of the one or more entity-type unified semantic objects. Its scope can be determined by specifying a root unified semantic object and taking the reachable closure along a preset relation type, explicitly specifying it according to the modeling task, or selecting it in reverse according to the semantic information requirements declared by the target projection. Completeness determination, projection generation, semantic conservation quantity comparison, and influence scope determination are all processed using the unified semantic projection scope as the unit; the case of only including a single entity-type unified semantic object is a special case. The unified semantic projection scope may include unified semantic objects whose object identities have not yet met the convergence criterion. These objects are only used to carry relational and constraint objects on them and do not participate in the current round of projection generation. Their identity identifiers are not included in the object identity set of the baseline semantic conservation quantity. After their identities converge, they are incrementally reprojected into the corresponding domain model.

[0024] Hard constraints and soft constraints: Hard constraints are those that, when violated, prevent the corresponding candidate semantic hypothesis or semantic change candidate from entering the determined semantic state. These include physical realizability constraints, geometric interference constraints, interface protocol compatibility constraints, and constraints explicitly marked as engineering-mandated. Soft constraints, when violated, only suppress the confidence of the corresponding candidate according to a preset penalty coefficient, without blocking its entry into the determined semantic state. Hard constraints are further divided into two categories based on their exemptibility: non-exempt hard constraints, which are explicitly marked as non-exempt, must be confirmed by deterministic evaluation in all cases, and have no exemption path; and exemptable hard constraints, which are required by default but can be exempted by an entity with corresponding authority according to explicitly configured exemption rules. When exempting, the exemption rule identifier, exemption entity, exemption reason, and scope of application must be recorded in the evolution field. User confirmation results participate in fusion as highly reliable semantic evidence, but do not exempt any hard constraints. Exemption can only be executed through the aforementioned explicit exemption rules and cannot be achieved through user confirmation or by increasing the evidence support. In this specification and claims, the term "unavoidable hard constraint" refers to the first category mentioned above; the term "hard constraint" without limitation refers to both categories mentioned above.

[0025] Projection Relationship Records: These are records established and stored during the projection generation of the domain model, representing the correspondence between unified semantic objects and domain model elements. They include at least: a unified semantic object identifier, a domain model element identifier, a projection instance handle, a projection operator identifier and version, a source semantic state version, a target model version, a mapping type, and references to the attributes, relationships, and constraints upon which the mapping depends. The mapping types include one-to-one, one-to-many, many-to-one, and many-to-many. Many-to-many correspondences are preferentially split into several independently maintainable one-to-one, one-to-many, or many-to-one records for separate storage, and are only recorded as many-to-many when reasonable splitting is not possible. Projection relationship records form the basis for reverse promotion, scope determination, and incremental reprojection. When the projection operator version of a target domain is upgraded, the existing projection relationship records for that target domain are marked as needing reconstruction and re-established during the next projection. Compatibility checks, failure markings, or reprojections are performed on projection relationship records and domain models affected by the version change. Projection relationship records and domain models of other target domains not affected by the version change remain unchanged. When establishing projection relationship records using attribute items as the mapping unit, the projection relationship record also records the attribute item identifiers mapped. The mapping type is determined based on the attribute item: one attribute item corresponding to one model element is recorded as one-to-one, one attribute item corresponding to multiple model elements is recorded as one-to-many, and multiple attribute items jointly determining one model element is recorded as many-to-one. It is preferentially split into several of the above records for separate storage; if it cannot be reasonably split, it is recorded as many-to-many. In the case where different attribute items of the same unified semantic object correspond to different model elements, the mapping type is determined separately for each attribute item, rather than being recorded as one-to-many simply because the object as a whole corresponds to multiple model elements. Otherwise, the determination benchmark for the mapping base will fluctuate between the object level and the attribute item level. Since reverse promotion and influence scope positioning are both performed on the attribute item level, inconsistent benchmarks will make patch positioning lose its determinism. All projection relationship records constitute a bidirectionally searchable projection dependency index: using the semantic element identifier or attribute item identifier as the key, all its projection instances and model elements can be retrieved; using the model element identifier as the key, the semantic elements, attribute items, and constraints it depends on can be retrieved. The determination of the scope of influence and the location of incremental reprojection patches are both performed based on this index. Furthermore, the generation of the draft state domain model also establishes mapping records, namely draft projection relationship records with draft state identifiers. In addition to the unified semantic object identifier or attribute item identifier, draft model element identifier, projection operator identifier and version, source semantic state version, and mapping type, its fields also include the draft projection instance handle, draft model version, and task version identifier. Draft projection relationship records are only used for locating draft model elements, reverse promotion of draft changes, and evidence tracing; they do not constitute formal projection relationship records, do not grant the draft model synchronization eligibility, and do not participate in the determination of the scope of influence and incremental reprojection for the formal domain model.The necessity of setting up this record lies in the fact that when the lift operator converts the draft model difference into semantic change candidates, it must answer three questions: "Which unified semantic object or attribute item does this draft model element correspond to?", "Based on which mapping is the reverse positioning performed?", and "Based on which draft model version is this modification?". However, the draft model has not obtained formal qualifications and does not have a formal projection relationship record. If a separate draft mapping is not established, the edits on the draft cannot be located and can only degenerate into coarse-grained lifts at the whole object level. Accordingly, the version validity of draft changes is determined based on the current draft model version and its task version identifier, not on the version of the formal domain model. The draft model and the formal domain model belong to two separate version sequences, and their version numbers are not comparable. If the draft changes are compared with the formal model version, all draft edits on the target projection that have not yet generated a formal model will be judged as outdated and unable to enter the state determination. The lifecycle of draft projection relationship records is subject to the following rules: When changes occur in the current task version involving the projection operator version, semantic information item structure, or target projection configuration, the original draft model and its draft projection relationship records are marked as to be rebuilt. Draft projection must be re-executed and the records rebuilt according to the new task version. Draft modifications that have not yet been submitted by the user are improved after performing a three-way comparison or rebasing on the regenerated draft model. Recalculating differences based solely on the old draft model is insufficient because, according to the definition of task version identifiers in the terminology section of this manual, the frozen content of the task version includes the projection operator version and semantic information item structure. After these changes, the composition, identifiers, and mapping types of the draft model elements may change, and the old records can no longer correctly locate the corresponding elements in the new draft. Once the target projection is officially generated and passes the conservation quantity verification, the corresponding draft model and its draft projection relationship record are marked as replaced or archived, and draft change events will no longer be accepted. If they are still allowed, the same target domain will simultaneously have two sources of input changes to the semantic base: the official domain model and the replaced old draft, resulting in version conflicts and duplicate entries by the same editor. Subsequent editing will create a new draft branch and a new draft projection relationship record based on the current official domain model or the current unified semantic projection range.

[0026] Tombstone State: The lifecycle state of a unified semantic object after it is determined to be deleted. Objects in the tombstone state retain their object identifier, evolutionary history, evidence citations, and deletion reasons, but do not participate in the current formal projection or cross-domain change propagation. Entering and exiting the tombstone state requires an evidence path and applies the same set of gating controls: An object enters the tombstone state only when a deletion semantic change candidate for that object is accepted and participates in evidence fusion, and its fusion result satisfies the convergence criterion and there are no unresolved hard constraint conflicts, thus forming an accepted semantic change. Similarly, an object reverts to a non-tombstone state only when a restoration semantic change candidate for that object is accepted and participates in evidence fusion, and its fusion result satisfies the same set of conditions, thus forming an accepted semantic change. The restoration event, restoration basis, and executing entity are recorded in the evolution field. If restoration is not gating-controlled, a deletion confirmed by sufficient evidence can be arbitrarily revoked by a low-reliability piece of evidence, and the tombstone mechanism loses its function of protecting semantic assets. Furthermore, two types of deletions must be distinguished: when the domain model or its model elements are deleted and promoted, it only indicates that the projection product of the target domain has been removed, and the corresponding unified semantic object remains valid; only when the deletion is promoted and accepted as a semantic deletion of the unified semantic object itself, does it enter the tombstone state. In addition, when an object enters the tombstone state, its reference integrity must be handled at the same time. For this purpose, the present invention defines two lifecycle states applicable to relational and constraint unified semantic objects. Failure state: applicable to relational unified semantic objects, it refers to the state in which the relation does not participate in the current formal projection or the reachable closure calculation of equation (13) because one or both endpoint objects of the relation enter the tombstone state or other unavailable state; the failure state only indicates that the relation is currently unavailable, and does not indicate that the relation itself has been semantically deleted. Its object identifier, evolution history and evidence tracing reference are all retained. Suspended evaluation state: applicable to constraint unified semantic objects, it refers to the state in which the constraint does not perform deterministic evaluation because the object it references is unavailable, and its satisfied state is not used as the basis for generation gating and propagation gating before evaluation is resumed. The entry into the above two states is directly triggered by the tombstoneing of the endpoint object or the referenced object and recorded in the evolution field without the need for further evidence fusion. This is because it only represents the objective fact that the object is "temporarily unavailable due to missing dependencies" and does not make a semantic judgment on the existence of the relationship or constraint itself. However, there is a strict order between setting the above states and determining the scope of influence: the change set must first be formed by the tombstone change and the identified unified semantic objects of various relation types and constraints. The scope of influence is determined by the union of the dependencies before and after the change, and the synchronization qualification of the affected formal domain model is temporarily revoked. Then, the object is set to an invalid state or a paused evaluation state, and then incremental reprojection is performed.The reason why this order cannot be reversed is that the relationship of the failure state no longer participates in the reachable closure calculation of equation (13) by definition. If the failure is set first and then the scope of influence is calculated according to the updated relationship, the connector, assembly relationship, dependency edge, constraint block and binding connector corresponding to the relationship in each domain model will no longer fall into the scope of influence. The incremental reprojection will only delete the model element corresponding to the object itself and omit the above-mentioned associated model elements. The dangling connection with missing endpoints and the constraint block of the object that has lost its function will remain in each domain model. Correspondingly, relations set to an invalid state and constraints set to a paused evaluation state retain their existing projection instance records in the projection dependency index, only used to generate deletion, disconnection, or deactivation patches, and are no longer used for subsequent ordinary semantic propagation. Correspondingly, when relational and constraint unified semantic objects enter a tombstone state, they still require the adoption of a deletion semantic change candidate for themselves and the fusion to form an adopted semantic change as a condition. They are not directly tombstoned due to the tombstoneing of endpoint objects. The difference is that the former can be restored with the restoration of endpoint objects, while the latter, once formed, must be restored through the same set of gating mechanisms for restoring semantic change candidates before exiting. Without the above processing, active relations pointing to tombstone objects and evaluable constraints referencing deleted objects will exist in the dynamic semantic state field for a long time, leading to three types of consequences: suspended model connectors or constraint blocks are generated during projection generation; the scope of influence calculation continues to propagate along invalid relations, causing incremental reprojection to touch target projections that should not be affected; and hard constraint evaluation uses the values ​​of deleted objects as input to obtain meaningless satisfied states. When an object is restored to a non-tombstone state, relationships in an invalid state and constraints in a paused evaluation state must be re-evaluated for validity and subject to the same gating as the object restoration. This means that the accepted state is obtained from its restoration semantic change candidates, and the accepted semantic changes are fused to form the conditions for restoration to an active state. This prevents relationships that became invalid during deletion from being indiscriminately revived during restoration. The restoration semantic change candidates are generated as follows: upon receiving a restoration event or restoration semantic input data for the object, corresponding restoration semantic change candidates are generated based on the preserved evolutionary history, evidence tracing references, the current state of the endpoint object, and the applicable constraint conditions. Then, state determination and fusion are performed according to the same rules as other semantic change candidates. If this generation method is not clearly defined, relationships in an invalid state and constraints in a paused evaluation state will remain indefinitely due to the lack of restoration candidates, and their associated structures will not be restored after the endpoint object is restored.

[0027] Evidence object anchoring, association threshold, and anchoring margin threshold: Evidence object anchoring refers to the process of determining which existing unified semantic object in the dynamic semantic state field each semantic evidence fragment in a newly arrived semantic evidence record should be associated with, or whether a new unified semantic object should be created. The association threshold is the absolute critical value for anchoring scores. The anchoring margin threshold is the critical value for determining the difference between the highest and second-highest anchoring scores. A definitive association is made only when both conditions are met simultaneously; if only the association threshold is met but the anchoring margin threshold is not, the semantic evidence fragment is retained as a candidate association. The anchoring result carries anchoring confidence and can be corrected by subsequently arriving semantic evidence.

[0028] Evidence fusion and convergence criterion: Evidence fusion refers to the process of incrementally updating the confidence of candidate semantic hypotheses using a probabilistic inference model, taking at least one of the semantic state, attributes, relationships, behavioral capabilities, and constraints of a unified semantic object as the variables to be estimated, and using the evidence support of semantic evidence fragments, pre-existing prior knowledge, and constraints as influencing factors. The convergence criterion refers to the conjunction of three conditions for determining whether a group of mutually exclusive candidate semantic hypotheses has been sufficiently determined, as detailed in Part (IV) of the invention. In this invention, the "determination conditions" of semantic content refer to the convergence criterion.

[0029] Replay-based refusion: When the association between semantic evidence fragments and unified semantic objects is corrected, evidence fusion is re-executed on each affected unified semantic object based on the relevant semantic evidence records and fragments stored in the evidence field in an append-only immutable manner, in order to reconstruct its semantic state, attributes, relationships, behavioral capabilities, constraints, and the confidence of candidate semantic hypotheses. After replay, the formal domain models that depend on the affected unified semantic objects are determined based on the projection relationship records. These models temporarily lose their synchronization eligibility because their semantic benchmarks have changed, and must be restored after re-determining completeness and passing the verification of projection and semantic conservation quantities. Other formal domain models that do not depend on the affected unified semantic objects are not affected.

[0030] Target projection set, semantic information items, and completeness: The target domain referred to in this invention is a generalized target model space, that is, a target model space defined by independently registered projection operators, lifting operators, and corresponding conservation quantity extraction logic; the "cross-domain" in this invention refers to cross-target model space, with cross-engineering domain being one of the main implementation scenarios. The situations in which two target domains differ include different engineering domains, different modeling language systems, different view types within the same modeling language system, different abstraction levels, and different execution environments; for example, the internal block diagram and state machine in the same SysML model constitute two target domains due to different view types, and the 3D scene model and scene semantic graph constitute two target domains due to different abstraction levels. Each registered independent projection operator, lifting operator, and conservation quantity extraction logic constitutes a different target domain as described in this invention, regardless of whether they belong to different industries or different modeling tools. The target projection set refers to the set of target projections selected in this round of tasks. Each target projection declares the semantic information items required to generate the corresponding domain model, the weights corresponding to each semantic information item, and the determined thresholds. The semantic information items are divided into mandatory semantic information items, optional semantic information items, and hard-constrained associated semantic information items. Completeness refers to the computable metric of the unified semantic projection range relative to a certain target projection, reflecting the degree of coverage of the semantic information requirements of the determined semantic information items for that target projection. Completeness is calculated separately relative to each target projection, and generation gating is performed on a per-target projection basis. Only when all target projections in the target projection set are required to be atomically generated simultaneously are they merged into a completeness relative to the target projection set according to a preset aggregation rule. Completeness is a task-relative attribute rather than an absolute attribute: the same unified semantic projection range may be complete for one target projection but incomplete for another.

[0031] Task Version Identifier: This is a version identifier generated for the target projection set and its configuration for this round of modeling tasks. The configuration includes the semantic information items, weights, determination thresholds, generation thresholds, and draft thresholds declared for each target projection, as well as the versions of the registered projection operators, registered boosting operators, conservation quantity extraction configuration versions, normalization rule versions, and deviation thresholds for each semantic conservation quantity component for each target domain. This task version remains unchanged during the completeness determination, projection generation, and consistency verification of this round. Newly generated target projection recommendations during execution are only written to the recommended target projection list. Changes to operator versions, conservation quantity extraction configuration versions, or normalization rule versions that occur during this period do not affect this round of tasks. Both are incorporated into the next task version after user confirmation or preset strategy judgment. The reason why the freezing scope is not limited to semantic information items and thresholds but must also cover the above versions is that if the plugin version changes during task execution, there will be a situation where completeness is calculated according to the old configuration, the domain model is generated according to the new projection operator, and semantic conservation quantities are extracted according to another version of extraction configuration and normalization rule. The results of the same task cannot be reproduced, and its consistency verification conclusions also lose meaning.

[0032] Active completion and acquisition actions: When the completeness of a target projection does not meet the preset model generation conditions, the system locates semantic information items under that target projection that have not yet met the convergence conditions, have not reached a certain threshold, or have unresolved conflicts. It then selects and executes an acquisition action from a set of candidate acquisition actions (including asking the user questions, requesting supplementary input data for a specified modality, reading a specified file, calling sensors, and calling external artificial intelligence models, etc.) based on the principle of maximizing the ratio of expected uncertainty reduction to acquisition cost. The supplementary input data is then used by the evidence generation module to regenerate semantic evidence records. The current round of completion stops when the preset termination conditions are met, and an incompleteness warning message indicating the semantic information items to be completed is output.

[0033] Projection operator (denoted as Π) d ): The mapping logic registered by the target domain d, which maps the unified semantic projection range to the domain model of the target domain.

[0034] Lifting operator (denoted as Λ) d): The mapping logic registered by the target domain d and used to generate semantic change candidates or semantic evidence records is the common carrier for reverse synchronization and import of existing models. It has two working modes: (1) Change promotion mode, which decomposes the model difference into one or more atomic model changes based on the domain change evidence records, the model difference before and after the domain model change, and the projection relationship records, and generates semantic change candidates by each atomic model change; (2) Initial import mode, which executes in two stages in the absence of the model before the change and the existing projection relationship records: The first stage is to parse the model elements, relationships, constraints and attribute items with object semantic content in the existing domain model file, and to obtain the semantic change candidates for each of them. Model elements, relations, constraints, and attribute items that can constitute independent semantic processing units generate semantic evidence fragments carrying independent positioning information, candidate semantic hypotheses, and source handles. Those that cannot constitute independent semantic processing units are carried along with the fragments of their respective model elements, thus forming semantic evidence records. Canvas coordinates, primitive styles, view collapse states, annotations, and editor auxiliary elements, which do not carry object semantic content, do not generate semantic evidence fragments. In the second stage, after each semantic evidence fragment has completed evidence object anchoring and has been associated with its corresponding unified semantic object, a temporary projection relationship record between the model element and the unified semantic object is established based on the model element source handle it carries. The order of these two stages is irreversible: the projection relationship record requires the unified semantic object identifier as a necessary field; no object identifier exists for recording before the object has been anchored and created. The temporary projection relationship record is converted into a formal projection relationship record after the relevant semantic information items meet the convergence criterion, have independent source verification (independent source semantic evidence, user confirmation, independent deterministic constraint verification, or the file is explicitly marked as a trusted initialization source; the mark must have both trust basis and integrity basis. The trust basis is one of the following: confirmation by a subject with preset permissions, trusted digital signature or source certificate, preset approval process, or pre-registered authoritative model library. The integrity basis includes at least content summary verification and version consistency verification, and the source record is complete. Those with only content summary and version consistency verification but no trust basis do not constitute a trusted initialization source, and the marked subject, basis, file content summary and confirmation time are recorded in the evolution field) and the semantic conservation quantity verification is passed.The semantic conservation quantity verification is performed in one of two ways: First, the expected semantic conservation quantity or verification expected model of the target domain is generated based on the updated unified semantic projection range. The existing domain model itself is used as the verification object for comparison without writing back. After passing, the existing domain model obtains the formal domain model state, and the temporary projection relationship record pointing to its model element is converted into a formal record. Second, the formal projection is re-executed based on the updated unified semantic projection range to generate the domain model of the target domain and a comparison is performed on it. After passing, the generated domain model is used as the formal domain model, and a formal projection relationship record is established separately for its model element. The temporary record pointing to the existing domain model element is retained as an import tracing relationship. If not all conditions are met, the existing model remains in the state of the imported baseline model. If the verification fails, it is handled according to the method used: If the first method fails, an additional failure event record is added, the existing domain model does not acquire the formal domain model state, and the temporary projection relationship record pointing to its model elements is marked as failing formalization and remains in a temporary state, which can be re-verified after supplementing evidence; If the second method fails, an additional failure event record is added, the generated model does not acquire the formal domain model state, the formal projection relationship record established for it is marked as invalid, and the temporary projection relationship record pointing to the existing model elements continues to be retained as the import source relationship; in both cases, the historical records are not physically deleted.

[0035] Conservative quantity extraction logic (denoted as Ext) d ): with the target field d The corresponding processing logic, which takes the generated domain model as input and re-extracts normalized semantic conservation quantities through the model parsing path of that domain, includes any of the following implementation forms: independently registered extraction functions, extraction branches that are independent of each other within the same domain plugin as the projection operator, and fixed extraction implementations for specific domains. All of these belong to the conservation quantity extraction logic described in this invention. Regardless of the implementation form used, the conservation quantity extraction does not reuse the intermediate results generated by the projection operator during the generation of the domain model.

[0036] Semantic conserved quantity and baseline semantic conserved quantity: A semantic conserved quantity (denoted as CQ) refers to the set of core semantic elements that must maintain consistency deviation within an allowable range during the projection of a unified semantic projection range onto any target domain. It includes at least one of the following: object identity identifier, unified topological relation structure, key parameters normalized to standard units, standardized interface signature, and at least one of the following standardized valid constraint sets; a baseline semantic conserved quantity (denoted as CQ) V() refers to the semantic conservation quantities directly calculated from the unified semantic projection range. Each target projection declares its set of mandatory conservation components, set of optional conservation components, deviation threshold for each component, and normalization rule version; the absence of components not declared as mandatory does not constitute a consistency check failure.

[0037] Formal domain models, draft domain models, and model synchronization eligibility status: Formal domain models refer to domain models that meet the semantic consistency condition through semantic conservation comparison and are marked as synchronizable, thus qualifying to participate in cross-domain change propagation. Draft domain models refer to domain models generated when their completeness does not meet the preset model generation condition but does meet the preset draft generation condition. They do not participate in semantic conservation comparison, are not marked as synchronizable formal domain models, and their content is not applied to any other domain models, but is provided for user preview and editing. However, when modifications to these models, after promotion and status determination, result in accepted semantic changes, these accepted semantic changes are equivalent to accepted semantic changes from other sources and are propagated to the affected target domains through a unified semantic status change event. Furthermore, domain models that still fail the semantic conservation comparison after the repair loop are marked as consistency verification failures; formal domain models whose semantic baseline changes due to replay-based re-fusion are marked as isolated versions. Neither of these two types of models is included in the automatic cross-domain synchronization link; their status is presented through the access and interaction layer. Furthermore, when a formal domain model receives an external change event, its eligibility for synchronization of the changed version is temporarily revoked until all semantic change candidates for this change have been processed and the semantic conservation quantity closed-loop verification has passed.

[0038] Model element isolation markers, source-side formal projection relationship records, and currently valid semantic views: Model element isolation markers refer to separating model elements, relationships, or attribute items that have not formed accepted semantic changes and carry object semantic content from the semantic content that the first formal domain model can currently participate in synchronization with. The isolation markers are implemented by marking the corresponding model elements, relationships, attribute items, or their projection relationship records as isolated states, and recording this isolation state in at least one of the domain model's accompanying metadata, isolated branches, or the evolution field of the dynamic semantic state field. The currently valid semantic view of the first formal domain model consists of model elements, relationships, attribute items, and constraints that are not in an isolated state, have currently valid formal projection relationship records, and exist as currently valid projection instances. The conservation quantity extraction logic uses the currently valid semantic view as the parsing object, excluding model elements, relationships, and attribute items that are in an isolated state or whose formal projection relationship records are in a state of pending reconstruction, termination, or invalidation during parsing. Model elements, relationships, and attributes in an isolated state do not participate in the current semantic conservation quantity extraction, cross-domain change propagation, or model difference benchmarks for subsequent formal domain model versions. Their original content, local differences, corresponding semantic change candidates, isolation reasons, isolation time, and historical projection relationship records are retained for supplementary evidence, user review, and auditing. For source model edits that have been accepted as semantic changes through evidence fusion, before performing closed-loop verification of the source domain model's semantic conservation quantity, formal projection relationship records on the source model side are established, updated, terminated, or marked as invalid according to the corresponding atomic model change type. New formal projection relationship records are established when new model elements, relationships, or constraints are added; their source semantic version and target model version are updated when existing model elements or attributes are updated; corresponding relational unified semantic objects and their projection relationship records are updated when relationships are reconnected; and corresponding projection relationship records are marked as invalid or terminated when projection products are deleted or semantic deletion occurs. Only after the formal projection relationship records have been processed can the corresponding accepted model content enter the current valid semantic view. When formal projection relationship record processing fails, the established accepted semantic changes are not rolled back, but the source formal domain model retains its synchronizability eligibility until the relevant projection relationship record is reconstructed and the closed-loop verification passes. When a semantic change candidate corresponding to an isolated model element, relationship, or attribute item is re-evaluated and becomes an accepted semantic change, the isolation is lifted according to the updated unified semantic projection range, and the corresponding formal projection relationship record is established, reconstructed, or activated. After passing the closed-loop verification of the source domain model semantic conservation quantity, it is included in the currently valid semantic view. When the corresponding semantic change candidate is rejected, a partial rollback or deletion is performed on the corresponding model element, relationship, or attribute item.Canvas coordinates, primitive styles, view collapse states, annotations, and editor auxiliary elements in the domain model that do not carry object semantic content are not model semantic content that needs to be supported by the currently determined semantic state. Their retention does not constitute a closed-loop check of the semantic conservation quantity of the source domain model.

[0039] Domain Change Evidence Record, Idempotent Event Index, and System Source Update: The domain change evidence record refers to an append-only immutable data record generated for a change event in the formal domain model that does not carry a valid system source update identifier and is confirmed by idempotency checks to be not a duplicate change event. It includes at least the change source, operating entity, change time, model difference before and after the change, projection instance handle, baseline semantic version identifier, baseline model version identifier, idempotency identifier, content digest for integrity verification, digital signature, message authentication code or other verification information, and a monotonically increasing write sequence number assigned by the dynamic semantic state field. Before generating the domain change evidence record and temporarily revoking the synchronization eligibility of the source formal domain model, the system first obtains or generates an idempotency identifier that uniquely represents the change event. When a change event carries a source event identifier allocated by its information source within the corresponding namespace and not reused by other change events, an idempotent identifier is generated based on the source identifier, target domain identifier, formal domain model identifier, and source event identifier. When a change event does not carry a source event identifier that meets the aforementioned conditions, an idempotent identifier is generated based on the source identifier, target domain identifier, formal domain model identifier, change operation type, local identifier or path identifier set of affected model elements or relationships, baseline model version identifier, and normalized change content summary. The identifier set is arranged in a preset order before participating in the generation. The operation subject, information source identifier, individual change object identifier, or individual change content summary cannot constitute an idempotent identifier on its own. The same idempotent identifier is generated when the same source event is transmitted repeatedly; different idempotent identifiers are generated for events acting on different formal domain models, different baseline model versions, or different model elements. The system then queries the indexes of idempotent events that are being processed and those that have already been processed. If the same idempotent identifier already exists in the index, the arriving event is determined to be a duplicate change event. The system generates a duplicate event reception record in an append-only immutable manner, recording the original source information, idempotent identifier, reception time, corresponding original change event identifier, and the duplicate relationship between the two. The duplicate event reception record does not have the identity of a new domain change evidence record, does not generate new semantic change candidates, does not participate in evidence fusion repeatedly, and does not revoke, restore, or change the current synchronizable eligibility status of the corresponding formal domain model due to the duplicate event. This eligibility status continues to be determined by the processing status of the original change event. Non-duplicate change events are registered as being processed and updated to being processed after their status determination, model-side handling, and source model eligibility processing are completed.System source update refers to the model update actually applied to the domain model by the computer modeling system through incremental reprojection. The system generates a system source update identifier for the system source update and stores this identifier in the evolution field along with the corresponding unified semantic state version identifier, the updated target model version identifier, and the update operation record. A valid system source update identifier is determined only when the system source update identifier and its associated information carried by the model update event are consistent with the internal records in the evolution field. Model update events carrying valid system source update identifiers do not generate new domain change evidence records, do not perform boosting operator processing or evidence fusion, in order to avoid synchronization echoes. System source update identifiers carried by external inputs but which cannot be matched with the internal records in the evolution field do not constitute valid system source update identifiers and are still processed as ordinary change events. The expected model or expected semantic conservation quantity used for closed-loop verification of the source domain model semantic conservation quantity is only an internal verification product of the system, is not written into the corresponding formal domain model, does not enter the domain model change event listening channel, and does not belong to the system source update. The baseline semantic version identifier is used to determine whether the change is based on an expired semantic state. The baseline model version identifier is used to determine whether the model difference is based on the current formal domain model version of the target domain. The idempotency identifier is used to prevent the same change from being repeatedly evidenced due to network retransmissions, message queue retries, or repeated notifications from modeling tools. The draft change evidence record and the domain change evidence record use the same record structure and are written to the evidence field in an append-only immutable manner. The projection instance handle in its field is taken from the draft projection instance handle, the baseline model version identifier is taken from the baseline draft model version identifier, and it also carries the task version identifier. The draft change event uses the same idempotency identifier generation and checking rules as the formal domain model change event, but its source model does not have synchronization eligibility, so it does not involve the revocation and restoration of synchronization eligibility. Its version validity is determined based on the current draft model version and the task version identifier.

[0040] Atomic model changes and semantic change candidates: An atomic model change refers to the smallest change unit corresponding to a single model element or the relationship between single model elements, obtained by decomposing the domain model difference based on the projection relationship record. A semantic change candidate refers to a semantic-level change proposal generated by a single atomic model change, involving one or more unified semantic objects, relation objects, or constraint objects within the unified semantic projection scope. After semantic change candidates are generated, they are in a pending evaluation state and are categorized into an accepted state, a candidate reserved state, a pending user confirmation state, a conflict state, or a rejection state based on the reliability of the source and the constraint verification results. The processing rules for each state are as follows: Semantic change candidates in the acceptance state are associated with the domain change evidence record as candidate semantic interpretations and participate in evidence fusion. Only when their fusion result satisfies the convergence criterion and there is no unresolved hard constraint conflict that results in an accepted semantic change, the fusion result is used to update the determined semantic state. Those that do not result in an accepted semantic change after fusion only update the confidence level of the corresponding candidate semantic hypothesis in the hypothesis field. Semantic change candidates in the candidate retention state also participate in evidence fusion, but their fusion result only updates the confidence level of the corresponding candidate semantic hypothesis in the hypothesis field and is not written into the determined semantic state. When their corresponding candidate semantic hypothesis satisfies the convergence criterion after subsequent evidence accumulation, the accepted semantic change corresponding to the subsequent evidence is written into the determined semantic state. Semantic change candidates in the pending user confirmation state do not update the determined semantic state for the time being. The user's confirmation or modification result participates in the fusion as a new semantic evidence record. Semantic change candidates in the conflict and rejection states do not participate in evidence fusion and do not update the unified semantic object. Their content and judgment basis are recorded in the evolution field for traceability.It should be clarified that "acceptance status" and "accepted semantic change" belong to two separate stages and should not be confused: The confirmation of a semantic change candidate as accepted only indicates that it has obtained the qualification to participate in evidence fusion. The acceptance conditions must include at least the following: the candidate has passed the preliminary validity check (i.e., the domain change evidence record on which it is based has passed the integrity check, its baseline semantic version identifier is consistent with the current semantic version, its baseline model version identifier is consistent with the version of the current formal domain model of the target domain or has completed rebasing and re-improvement, and is not a repeated idempotent event); the reliability of the change source is not lower than a preset threshold; the initial evidence support is not lower than a preset threshold; there are no semantic ambiguities that cannot be automatically resolved; and it does not violate the established non-exemptible hard contract. The constraints and the exemptable hard constraints involved are satisfied or have been effectively exempted. Those that fail the pre-validity check are marked as integrity anomalies or pending rebasing, falling neither into the accepted state nor the candidate retention state, and do not participate in evidence fusion. The purpose of including "no semantic ambiguity that cannot be automatically resolved" in the acceptance condition is to ensure that candidates with ambiguity, even if their source reliability and support meet the standards, do not fall into the accepted state, but instead enter the pending user confirmation state according to the diversion priority, thus maintaining consistency with the processing of two mutually exclusive candidates generated by deletion-type atomic model changes that must be decided by the user. After fusion, if the fusion result of the candidate satisfies the convergence criterion and there is no conflict of unresolved hard constraints, it constitutes an accepted semantic change and is written into the determined semantic state. Semantic change candidates that have obtained the accepted state but do not satisfy the convergence criterion after fusion only update the confidence of the corresponding candidate semantic hypothesis in the hypothesis field, are not written into the determined semantic state, and do not trigger the determination of the scope of influence and incremental reprojection. The distinction between the accepted status and the candidate reservation status lies in whether the above-mentioned acceptance conditions are met: the former has already met the fusion access qualification, while the latter has not reached the threshold and only accumulates confidence in the form of candidate hypotheses.

[0041] Scope of Influence and Incremental Reprojection: Scope of influence refers to the set of semantic elements and target domain projections affected by a certain change event, determined by the relationships and constraints in the updated unified semantic projection scope. Incremental reprojection refers to a local projection re-executed only for the affected target domain projections within the scope of influence, based on the updated unified semantic projection scope. In this invention, the initial projection generation of the domain model, the generation of the draft state domain model, and the incremental reprojection are all uniformly executed by the projection and consistency verification module. Both the formal domain model and the incremental reprojection product must pass semantic conservation quantity verification before obtaining synchronization eligibility; there is no model generation path that bypasses semantic conservation quantity verification to obtain synchronization eligibility. The draft state domain model does not participate in semantic conservation quantity comparison, nor does it obtain synchronization eligibility; it is only presented as a draft state after passing structural validity checks and non-exemptible hard constraint checks.

[0042] Formal bidirectional index: A bidirectional searchable correspondence between semantic evidence fragments and accepted semantic conclusions; it allows tracing back from accepted semantic conclusions to all semantic evidence fragments supporting those conclusions and their associated semantic evidence records, and also allows locating all accepted semantic conclusions supported by semantic evidence fragments.

[0043] Restricted system candidates: Candidate content generated internally by the computer modeling system that has not yet obtained independent external semantic evidence support, user confirmation, or independent deterministic verification. Restricted system candidates include: historical prediction candidates generated based on the evolutionary history of a unified semantic object to describe possible future semantic states; correction suggestion candidates generated when hard constraints are determined to be violated or parameters do not meet constraints; and repair proposal candidates generated after the deviation of semantic conservation quantities is located. The generation mechanisms, triggering links, and technical functions of the above three types of candidates are different and are not confused with each other, but they are all subject to the restricted participation rules described in this specification. Restricted system candidates do not have the identity of semantic evidence records, are not written into the evidence field, and are not included in the number of evidence items, independent evidence related groups, or semantic evidence source categories in evidence fusion. Instead, they are stored in the hypothesis field or evolution field as candidate auxiliary factors independent of the semantic evidence likelihood term, and participate in the confidence calculation of the corresponding candidate semantic hypothesis with restricted weights. Each restricted system candidate includes at least a candidate identifier, candidate category, generator identifier and version, baseline semantic version identifier, target semantic information item, generation basis, derivation chain identifier, related group identifier, idempotency identifier, generation time, valid status, and failure reason. For restricted system candidates with the same baseline semantic version, the same generator version, the same target semantic information item, and semantic equivalence, the system performs replacement, merging, or discounted weighting, and does not repeatedly accumulate them as mutually independent candidates; the generation or updating of restricted system candidates is not included in the number of updates of independent evidence-related groups used in the convergence determination. After the restricted system candidate is confirmed by the user, solved independently, simulated, verified deterministically, or supported by an external information source, a new semantic evidence record is generated. The newly generated semantic evidence record is used to record the restricted system candidate identifier on which it is based, while the original restricted system candidate continues to be retained as a generation process record. Semantic fusion configuration version: used to identify the version information of the fusion rules and parameter set used in a certain evidence fusion, replay-based refusion, or semantic state re-evaluation. The semantic fusion configuration version is associated with at least the following: the evidence object anchoring algorithm identifier, version, and threshold; the evidence fusion algorithm identifier and version; the pre-set prior knowledge version; the evidence source reliability calculation rule version and the snapshot of the source reliability weights used in this fusion; the constraint rule version; the convergence threshold, confidence difference threshold, and number of stable updates; the conflict arbitration rule; the evidence-related group division and discount weighting rule; and the rules for generating, deduplicating, invalidating, and weighting restricted system candidates. Each time a determined semantic state is formed, the supporting evidence set identifier, the semantic fusion configuration version, and the fusion result version are recorded in the evolution field to support the reproduction of historical results and re-evaluation according to the current rules.

[0044] A confirmed semantic withdrawal event is an append-type immutable event used to record a semantic information item that was originally in a confirmed state, but whose supporting results no longer meet the convergence condition, corresponding determination threshold, or hard constraint gating condition due to changes in evidence fragment anchors, evidence invalidation, correlation modification, changes in constraint rules, or re-evaluation according to the current configuration. Therefore, the semantic information item needs to be changed from its current confirmed semantic state to a candidate semantic hypothesis state, an undetermined state, or a conflicting state. The confirmed semantic withdrawal event at least records the identifier of the withdrawn semantic information item, the original confirmed semantic content and its version, the reason for withdrawal, the triggering evidence or related event, the semantic fusion configuration version used for replay or re-evaluation, the candidate semantic hypothesis after withdrawal and its confidence level, the scope of impact, and the processing time. The withdrawal does not delete or rewrite the original confirmed semantic state and its historical versions; instead, it generates a new unified semantic object version so that the withdrawn state becomes the current valid state.

[0045] Access and Interaction Layer: Located at the system boundary, this layer is responsible for input adaptation and result presentation. It includes input adapter groups, question-and-answer and confirmation channels, and domain model presentation and editor. The part of the user output used to determine the semantic content of the object is processed by the evidence generation module to form semantic evidence records. The part used to set the target projection set, threshold, permissions and presentation mode is recorded as control configuration input in the evolution field and does not participate in semantic fusion.

[0046] The technical problems to be solved by this invention include: First, how to automatically, traceably, and error-correctably restore input data from heterogeneous information sources (including external artificial intelligence models) into a unified semantic object; Second, how to computationally determine the completeness of the unified semantic object relative to the modeling target, and actively complete or downgrade the output when it is incomplete, so as to avoid the outflow of incomplete models; Third, how to ensure the semantic consistency of the same semantic base after projection onto multiple target domains, so that only the domain models that pass the verification are qualified for synchronization; Fourth, how to ensure that changes to any domain model are safely propagated to other domains through evidence-based backflow, without relying on point-to-point direct mapping between domain models and without generating synchronous echoes.

[0047] To address the aforementioned technical problems, the first aspect of this invention provides a cross-domain modeling method for unified semantic objects, applied to a computer modeling system. The computer modeling system registers projection operators and lifting operators for at least two target domains, and configures conservation quantity extraction logic for each target domain. The method includes steps S1 to S5. The input received by the computer modeling system is divided into two categories: semantic input data and control configuration input. Semantic input data is used to determine semantic content and is always entered through semantic evidence records. Control configuration input is used to set operating parameters such as thresholds, weights, and target projection sets, without going through an evidence layer.

[0048] S1. Obtain input data from one or more information sources, and generate a semantic evidence record based on the semantic input data used to determine the semantic content of the object. The semantic evidence record contains one or more semantic evidence fragments, each of which has independent location information, one or more candidate semantic hypotheses, and initial evidence support corresponding to each candidate semantic hypothesis. The semantic evidence record also includes source information and a timestamp. The semantic evidence record is stored in a dynamic semantic state field in an append-only immutable manner. The semantic input data from any information source participates in the update of the unified semantic object in the form of a semantic evidence record, without directly overwriting the determined semantic state, relationship, or constraint of the unified semantic object.

[0049] S2. Using semantic evidence fragments as units, perform association matching with existing unified semantic objects in the dynamic semantic state field. Based on the association matching results, associate the semantic evidence fragment with any existing unified semantic object of any type, or create a new unified semantic object and establish an association based on the semantic element type indicated by the semantic evidence fragment. When the fragment indicates an independently referential object, create an entity-type unified semantic object; when it indicates a typed relationship between two or more unified semantic objects, create a relation-type unified semantic object; when it indicates a assessable constraint expression acting on one or more unified semantic objects, create a constraint-type unified semantic object. Merge one or more semantic evidence fragments associated with the same unified semantic object, update at least one of the semantic state, attributes, relationships, behavioral capabilities, and constraints of the unified semantic object, and retain semantic content that has not yet met the convergence conditions as a candidate semantic hypothesis state without writing it into the determined semantic state.

[0050] S3. Obtain the target projection set corresponding to this round of modeling tasks. The target projection set includes target projections corresponding to one or more registered target domains. Determine the unified semantic projection range of this round of modeling tasks. The unified semantic projection range includes one or more entity-type unified semantic objects, and at least one of the following: entity-type unified semantic objects associated with them, and constraint-type unified semantic objects acting on them. When no entity-type unified semantic objects or constraint-type unified semantic objects exist, the range is composed of the one or more entity-type unified semantic objects, and serves as the processing unit for completeness determination, projection generation, semantic conservation quantity comparison, and influence range determination. Based on the semantic information requirements corresponding to the target projection set, determine the unified semantic projection range. The completeness of the unified semantic projection range relative to each target projection; when the completeness meets the preset model generation conditions, the unified semantic projection range is projected to generate a domain model corresponding to one or more target projections in the target projection set through the projection operator registered in the corresponding target domain, and a projection relationship record between the semantic elements in the unified semantic projection range and the model elements in the generated domain model is created and stored. The domain models corresponding to the at least two different target domains are generated by executing this step once or multiple times; when the completeness does not meet the preset model generation conditions, the generation of the formal domain model in this round is stopped, or at least one of the active completion process or the draft state domain model generation process is entered, and the corresponding final state is presented through the interaction layer.

[0051] S4. Obtain a baseline semantic conservation quantity from the unified semantic projection range. The baseline semantic conservation quantity includes at least one of the following: object identity identifier, topological relationship structure, key parameters, interface signature, and effective constraint set. Then, through the conservation quantity extraction logic set corresponding to each target domain, normalized semantic conservation quantities are obtained by taking the generated domain model as input. The consistency deviation between the two is compared. Only the domain model that meets the semantic consistency condition is marked as a synchronizeable formal domain model. Those that do not meet the condition enter the repair loop or are marked as consistency verification failure and presented through the interaction layer.

[0052] S5. Upon receiving a change event for a synchronizeable first formal domain model in the first target domain, an appended immutable domain change evidence record is generated based on the change event. The synchronizeability of the changed first formal domain model is temporarily revoked. Using the promotion operator registered in the first target domain, the model difference before and after the change is decomposed into one or more atomic model changes based on the projection relationship record. Each atomic model change generates semantic change candidates involving one or more unified semantic objects, relation objects, or constraint objects within the unified semantic projection range. A state determination and corresponding processing are performed on each semantic change candidate. Semantic change candidates with an accepted state are associated with the candidate semantic interpretation. Domain change evidence records and participate in evidence fusion. Only when the fusion result satisfies the convergence condition and there are no unresolved hard constraint conflicts resulting in an accepted semantic change is the fusion result used to update the determined semantic state. Semantic change candidates that have not obtained an accepted state, and semantic change candidates that have obtained an accepted state but have not formed an accepted semantic change after fusion, do not update the determined semantic state. For details on the specific set of candidate states, decision priorities, handling of unaccepted cases, and closed-loop verification and echo suppression, please refer to Part (VII). Only when the semantic change candidate forms an accepted semantic change that satisfies the convergence condition and has no unresolved hard constraint conflicts after evidence fusion, is the updated semantic change determined based on the relationships and constraints in the unified semantic projection range. The scope of the change event; when one or more second target domain projections are affected within the scope of the change, based on the updated unified semantic projection scope and the projection operators registered by each second target domain, the synchronization eligibility of the corresponding formal domain models of the affected second target domains is temporarily revoked. Incremental reprojection is then performed on each affected second target domain projection, and synchronization eligibility is restored only after the incremental reprojection is completed and the resulting domain model passes the semantic conservation comparison. Synchronization eligibility is restored after all semantic change candidates related to this change have completed their status determination and corresponding processing, and the changed first formal domain model passes the semantic conservation closed-loop verification. Furthermore, the determination of the scope of the change... In addition to being triggered by domain model change events, incremental reprojection can also be triggered by unified semantic state change events: when semantic evidence records generated from semantic input data from any information source are fused to form accepted semantic changes, causing changes in the determined semantic content within the unified semantic projection range, and these changes are not caused by changes in the synchronizable formal domain model, a unified semantic state change event is generated. Based on this, the scope of influence is determined, and the synchronizable qualification of the corresponding formal domain model for each affected target projection is temporarily revoked. Then, incremental reprojection is performed, and the qualification is restored only after the semantic conservation quantity comparison is passed. In this case, the semantic conservation quantity closed-loop verification of the source domain model is not performed. See Part (VII) for details.Specifically, cross-domain change propagation is performed via a unified semantic projection range and projection and promotion operators within a dynamic semantic state field, rather than directly applying changes from the first formal domain model to the domain model of the second target domain.

[0053] A second aspect of this invention provides a cross-domain modeling system for unified semantic objects, including at least one processor and a memory. The system registers projection operators and boosting operators for at least two target domains, and configures conservation quantity extraction logic for each target domain. The system implements an evidence generation module, a semantic object association and fusion module, a completeness determination module, a projection and consistency verification module, a change evidence and boosting module, and an impact analysis and incremental reprojection module. The impact analysis and incremental reprojection module sends an incremental reprojection request to the projection and consistency verification module after determining the scope of influence. The initial projection generation of the domain model, the generation of the draft domain model, and the incremental reprojection are all uniformly executed by the projection and consistency verification module. The projection and consistency verification module is also used to isolate domain models that have not obtained or have had their synchronization eligibility revoked from the automatic cross-domain synchronization link and present their state through the interaction layer. When the change evidence and boosting module receives an external change event of the formal domain model, it temporarily revokes the synchronization eligibility of the changed model and restores the eligibility after the relevant semantic change candidates have been processed and the semantic conservation quantity closed-loop verification has passed. A third aspect of the present invention provides a computer-readable storage medium on which a computer program stored is executed by a processor to implement the above-described method. A fourth aspect of the present invention provides a computer program product comprising a computer program that, when executed by a processor, causes the processor to implement the above-described method.

[0054] The core components and mechanisms of this invention are described in detail below. To avoid ambiguity, each symbol in this specification represents only one meaning, and different symbols are used for different meanings, without distinguishing between different meanings by font, font shape, or capitalization. The symbols used are conventionally as follows: U Represents a unified semantic object. U * Represents the true semantics of the real-world object to be modeled. V Indicates the scope of the unified semantic projection; b U 、Rel U Att U Cap U Con U ,Trc U Hst U 、Prj U The following are represented sequentially: the confidence function, relational structure, attribute function, set of behavioral capabilities, set of constraints, set of evidence citations, evolutionary history, and projection relationship of the unified semantic object; Ont Represents a hierarchical set of semantic concepts. h Indicate candidate semantic hypotheses; e Represents semantic evidence records, f Represents semantic evidence fragments, ℓ f Indicates the spatiotemporal location of a segment. s k Indicates the initial level of evidence support. q e Indicates the quality score of the evidence; z Ovl represents the anchoring score, and Ovl represents the spatiotemporal localization overlap. ϕ Represents feature mapping, G Indicates the contextual consistency of the relationship; v Represents semantic information items, s d This represents the set of semantic information items declared by the target projection d. w v The weights of semantic information items, i v det This indicates its defined threshold. b v Indicates its confidence level; d This represents the target projection and its corresponding target region. G Represents the target projection set, Cmp d With Cmp G Gen represents subcompleteness and set completeness, respectively. d Indicates the formal generation condition indication, Out d This represents the decision results of the hierarchical generation gating, P1, P2, P3. These represent the three gating processes in sequence: formal projection, draft generation, and active completion. act Indicates the data collection action. Act Represents the set of candidate collection actions. Unc CQS represents uncertainty. This indicates a shared, conserved space. CQ Represents a semantically conserved vector. r Indicates the category number of the conserved quantity component. d d,r , l d,r , D d The components are represented in order: normalization bias, component bias weights, and overall consistency bias; П d Λ d Ext d , M d The projection operator, lifting operator, conservation extraction logic, and domain model of the target domain d are represented sequentially; Δ V Represents the unified semantic projection range change set, Cl This represents the reachability closure on the relation field and the constraint field. Imp Indicates the scope of influence. or Representing semantic elements, p Prj represents a projection instance. V Represents the set of projection instances within the scope of the unified semantic projection; x Lap Inc. These represent, in turn, the conservation vectors assigned to the layer structure, the layer Laplace matrix, and the inconsistency with the global structure; n D This indicates the number of target domains. Each threshold and coefficient is expressed as follows: i , t , c , r and its subscript indicates, where i anc , i mar , i conf , t 1 , t 2 , n stab , c gen,d , c draft,d , i d cons , i d,r cons , i glob , r The following are the threshold values, respectively: association threshold, anchoring margin threshold, conflict threshold, first convergence criterion threshold, second convergence criterion threshold, number of stable windows, generation threshold, draft threshold, total consistency deviation threshold, individual consistency deviation threshold, global consistency threshold, and restricted decay coefficient. Symbols not listed in this table are defined at their first appearance and also follow the convention that each symbol represents only one meaning.

[0055] (a) Unified semantic objects and dynamic semantic state fields.

[0056] The unified semantic object U is formally defined as an octet, as shown in equation (1): (1)

[0057] In formula (1): b U For the confidence function of this object, any candidate semantic hypothesis for this object. h Give its current confidence level. b U ( h The candidate semantic hypotheses ∈ [0,1] include concept class hypotheses, attribute value hypotheses, and relation hypotheses, where the concept class hypotheses are taken from an expandable hierarchical set of semantic concepts. Ont Non-mutually exclusive hypotheses with hierarchical inclusion relationships can simultaneously have non-zero confidence. Mutually exclusive hypotheses at the same level constitute a group of mutually exclusive candidate semantic hypotheses, and the confidence of the group is normalized. Rel U A typed relational structure is a set of directed typed edges between the object and other unified semantic objects. Its elements can be either inline edges held by the object itself or references to already materialized relational unified semantic objects. Att U This is a property function, where the property value is related to time. t With context ctx function Att U ( t,ctx (and include the unit of measurement, value distribution, and evidence cited.) Cap U It is a set of behavioral capabilities. Con U For a set of constraints, constraints are evaluable conditional expressions, which are divided into hard constraints that cannot be exempted and soft constraints that only suppress confidence. Their elements can be either inline constraints held by the object itself or references to an already materialized constraint-type unified semantic object. Trc U This is a collection of citations for evidence tracing. Hst U For evolutionary history. Prj U This refers to the projection relation, that is, the set of projection instance handles of the object in each target domain. The eight components of Equation (1) are equally applicable to the three types of unified semantic objects: entity type, relation type, and constraint type. Rel U Record the type associations between it and each endpoint object, Att U The attributes that carry the relation itself; a constraint-type unified semantic object b U Characterizing the confidence level of each candidate value of its constraint expression, Con U Other constraints referenced by this constraint.

[0058] The core technical stance of this solution is: the domain model is not the object itself, but a projection of the object; the input data is not the source of the model, but an observation of real-world objects. Let... U * For the true semantics of the real-world object to be modeled, the j-th input satisfies the observation model, as shown in equation (2): (2) In equation (2): y j Input data for the j-th channel; U * The true semantics of the real-world object to be modeled; c j This includes the type, modality, task, and context of the information source; p j This is the observation generation model corresponding to the information source, that is, the conditional probability distribution that generates the input under given true semantics and context. Equation (2) adopts the conditional distribution form instead of the fixed function noise form to uniformly cover continuous numerical inputs and discrete, structured inputs: for continuous measurements such as geometric dimensions, temperature, and power consumption, Equation (2) degenerates into additive noise form. y j = g j ( U * )+ e j ,in g j For observation function, e j This is a special case for observing noise; for inputs such as text, program code, graph structures, interface types, discrete relations, image recognition candidate categories and constraint expressions, pj It manifests as the conditional distribution of the true semantics when they are expressed, recognized, or recorded in this modality. Its deviation manifests as missed detections, false detections, referential ambiguity, type confusion, and expression omissions, rather than numerical noise superimposed on the values. The essence of modeling is thus reconstructed as "recovering a unified semantic object from multiple observations, and then generating domain models from the unified semantic object". The fusion of Equation (5) is the computable realization of this recovery process.

[0059] The dynamic semantic state field is a global data structure that carries all unified semantic objects. It consists of six sub-fields: entity field, relation field, constraint field, evidence field, evolution field, and hypothesis field. The dynamic semantic state field provides transactional read / write, subscription notification, and snapshot query capabilities to each functional module. It is the sole medium for exchanging semantic content and business state data between modules: all semantic content required by any module is read from the dynamic semantic state field, and all semantic content produced is written into the dynamic semantic state field. Modules do not directly exchange semantic content and business state data, thus forming a star topology centered on the dynamic semantic state field. Modules can directly transmit control signals without carrying semantic content. For example, the completeness determination module sends a projection trigger signal to the projection and consistency verification module after determining that a target projection meets the preset model generation conditions; the impact analysis and incremental reprojection module sends an incremental reprojection request to the projection and consistency verification module after determining the scope of influence. These signals only contain event identifiers, task version identifiers, and semantic element identifiers. The receiving module uses these to read the corresponding semantic content from the dynamic semantic state field before processing, thus not disrupting the aforementioned star topology. Unified semantic objects persist independently of any domain model within the dynamic semantic state field; even after any domain model is created, modified, or deleted, its associated unified semantic objects, along with their evidence tracing and evolutionary history, remain available. The unified semantic projection scope V is a subset of semantics relevant to the current modeling task within the dynamic semantic state field, consisting of one or more entity-type unified semantic objects, and at least one of the associated entity-type relational unified semantic objects and constraint-type unified semantic objects acting upon them; in the absence of relational and constraint-type unified semantic objects, ... V It consists of one or more entity-type unified semantic objects, which is a special case of it.

[0060] (ii) Semantic evidence records and additional immutable evidence layers.

[0061] The first core design difference between this scheme and the "direct input generation model" type of technology is the introduction of an intermediate evidence layer. The definitions of semantic evidence record e and its semantic evidence fragment f are shown in equation (3): (3) In equation (3): ide Used as a unique identifier for evidence; src e Source information (source type, source identifier, extractor identifier, and version); t e For timestamps; q e ∈[0,1] represents the overall quality score of the evidence; F e This is the set of semantic evidence fragments contained in the record. n e Number of segments; id f Used as a unique identifier for a segment; ℓ f For the independent spatiotemporal location of this segment (image region coordinates, video frame intervals, document paragraphs, canvas coordinates, etc.); Oh f This is the set of candidate semantic hypotheses for this segment. h k For the first k One candidate semantic hypothesis, s k ∈[0,1] is its initial evidence support. n f This represents the number of candidate hypotheses for the segment. By using segments rather than the entire record to carry the location and candidate hypotheses, an image, requirement document, or existing model file that describes multiple objects can be anchored to different unified semantic objects, and the association of any segment can be modified individually without affecting the other segments.

[0062] The evidence field stores semantic evidence records in an append-only, immutable manner. Immutability is not a choice of storage style, but rather a functional prerequisite for several mechanisms in this invention: First, anchoring correction and replay-based re-fusion require a logically unmodifiable and undeletable original evidence benchmark. If the evidence can be rewritten, the object state cannot be reconstructed from the credible benchmark after the association is corrected. Second, the traceability validity of the formal bidirectional index requires that the evidence content be strictly consistent with the original conclusion. Third, concurrent changes are merged according to the evidence's chronological order, requiring that the record's temporal sequence not be rewritten. Without immutability, anchoring errors become irreversible and destructive operations, rendering the audit chain ineffective. This is the essential difference between this invention and an overwriteable knowledge base: a knowledge base updates by rewriting records, while this invention evolves by adding evidence and reinterpreting; the conclusion may change, but the observation remains.

[0063] The evidence layer decouples "generation" from "acceptance": the semantic input data (including the reasoning results of external artificial intelligence models) in the output of any information source is only evidence, and whether it enters the determined semantic state is determined by subsequent fusion convergence, constraint verification and gating.

[0064] (iii) Anchoring of evidence objects.

[0065] When a new semantic evidence record arrives, the semantic object association and fusion module performs evidence object anchoring on each semantic evidence fragment f in that record. The anchoring score of fragment f relative to the unified semantic object U is defined in equation (4): (4) In equation (4): Ovl( ℓ f ,ℓ u The spatiotemporal overlap is represented by the range [0,1]. ℓ u To unify semantic objects U Known spatiotemporal location; ϕ (·) To map semantic evidence fragments and unified semantic objects to a shared semantic feature space, sim(·,·)∈[0,1] is the similarity measure in this space (preferably normalized cosine similarity). C ( f,U The relational context consistency is defined as ∈ [0,1], which is calculated based on the degree of matching between the relational information indicated by the fragment and the relational structure established by the fusion of prior evidence in the relational field. m 1 , m 2 , m 3 ≥0 is the weight coefficient and m 1 + m 2 + m 3 =1, so that z ( f,U )∈[0,1].

[0066] When the highest anchoring score is not lower than the preset association threshold i anc Furthermore, the difference between the highest anchoring score and the second highest anchoring score is not less than the preset anchoring margin threshold. i mar When the highest anchoring score is not lower than a certain threshold, the semantic evidence fragment is associated with the corresponding unified semantic object; i ancHowever, the difference is lower than i mar When there are two or more indistinguishable candidate objects, no definitive association is made. Instead, the semantic evidence fragment is retained as a candidate association, awaiting further evidence or user confirmation before a definitive association is established. This avoids prematurely solidifying the association before the object boundaries are clearly defined. When there are no existing objects or the highest score is lower than [a certain threshold], [further action is taken]. i anc At that time, a new unified semantic object is created and an association is established. The margin condition is isomorphic to the difference margin condition in the convergence criterion of equation (6): the absolute threshold answers "whether it is similar enough", and the difference margin answers "whether it is separable enough", and neither can be omitted. It should be emphasized that in equation (4) C ( f,U The value of ) depends on the result of the prior fusion and writing into the relation field, while the input of fusion depends on the output of anchoring. Anchoring and fusion constitute a self-correcting cyclic coupling that evolves with the dynamic semantic state field, unlike one-time entity matching: the anchoring result carries anchoring confidence, which can be corrected by subsequent evidence. When correcting, it triggers the replay-based refusion described in Part (VIII). If this cyclic coupling is missing, early erroneous associations will be permanently solidified and contaminate all subsequent fusions. When some components of spatiotemporal location overlap, semantic feature similarity, and relational context consistency cannot be obtained due to input data type limitations, the missing components can be ignored, a default value can be set for the missing components, or the weights of the remaining components can be re-normalized before calculating the anchoring score. When there is only one anchoring score not lower than in the dynamic semantic state field i anc When there is no second-highest anchor score for a unified semantic object, the anchor margin condition is waived, or the second-highest anchor score is taken as a preset baseline value before calculating the difference. This waiver is isomorphic to the convergence criterion of Equation (6) when there is only a single valid candidate in the mutually exclusive candidate group, and the two together avoid the decision being suspended because the margin cannot be calculated when there is no comparable object. The semantic evidence fragments retained by the candidate association relationship also follow the evidence entry discipline for subsequent association determination: the user's confirmation result of the fragment's ownership is recorded as a new semantic evidence record, or written as an appended association confirmation event into the dynamic semantic state field. After the association state is updated, a definite association is formed. The historical candidate association records are not deleted or rewritten, so that every change in the association can be traced back.

[0067] (iv) Evidence fusion, convergence criteria and conflict arbitration.

[0068] Evidence fusion uses at least one of the following—semantic state, attributes, relationships, behavioral capabilities, and constraints—of a unified semantic object as variables to be estimated, and establishes a probabilistic inference model using evidence support, pre-existing prior knowledge, and constraints as influencing factors. For a set of mutually exclusive candidate semantic hypotheses for a given semantic information item to be decided, any candidate hypothesis after fusion...h The confidence level is expressed as a log-linear pooling weighted by source reliability, as shown in equation (5): (5) In equation (5): b U ( h ) represents the confidence level of the candidate hypothesis h after fusion, which is normalized within the mutually exclusive candidate group. The result is the confidence function written into equation (1). π 0 ( h The prior confidence level given for pre-defined prior knowledge; N e The number of pieces of evidence involved in the fusion; L ( e i | h ) is the first i The support of a piece of evidence for hypothesis h is determined by the set of candidate hypotheses for the corresponding segment of that evidence. Oh f The corresponding initial evidence support s k With the quality of evidence Jointly determined; α i ≥0 is the first i The reliability index of the source of evidence is weighted, and the reliability of the source can be configured according to the type of information source and dynamically adjusted according to the historical accuracy. N c The number of constraint factors. ψ c ( h )∈[0,1] is the first cConstraint factors, and the assumptions that violate the physical, geometric or logical constraints in the constraint field are suppressed accordingly. Engineering implementation of logarithmic domain incremental update on the optimal factor graph: when new evidence arrives, only the affected estimated variables are locally updated without full graph recalculation, supporting real-time evolution under streaming inputs such as video streams and sensor streams. Evidence independence processing. The product form of Equation (5) is based on the premise that each piece of evidence is independent of each other. However, in actual input, the same original input is often processed by multiple parsers, the same external artificial intelligence model outputs approximate results multiple times for the same input, or the same file is imported multiple times. Such semantic evidence fragments are not independent of each other. Therefore, each semantic evidence record carries an original source identifier and a derivation chain identifier. Semantic evidence fragments from the same original input or the same derivation chain are classified into the same evidence related group; fragments within the same related group are not counted repeatedly in Equation (5) as mutually independent evidence, but are weighted by combining according to a preset discount coefficient, or only the one with the highest source reliability is weighted; the number of stable windows in Equation (6) n stab It is also calculated based on the number of updates of the relevant groups of independent evidence, rather than the number of evidence items. Without this processing, the repeated superposition of evidence from the same source can push the same candidate hypothesis past the convergence criterion and make it meet the stability window condition, forming an unstable path similar to the "predictive self-reinforcement" described in Part (X).

[0069] Convergence Criterion. Let the confidence levels of the highest and second-highest candidate hypotheses in a set of mutually exclusive candidate semantic hypotheses be respectively... b (1) and b (2) The group is considered convergent if and only if all three conditions of equation (6) are satisfied simultaneously: (6) In equation (6): t 1 This is the first threshold (absolute confidence condition). t 2 This is the second threshold (difference margin condition). stab ( n stab ) is a stable window indicator function if and only if the most recent consecutive n stab In each evidence update, the candidate hypothesis with the highest confidence level remains unchanged and is set to 1. n stab This is the preset number of stable windows. The difference condition is waived when only one valid candidate exists in the mutually exclusive candidate group.

[0070] Each of the three conditions has a clear technical effect: the first threshold prevents hasty conclusions when evidence is insufficient; the second threshold prevents misjudgments when two candidates are close, as the absolute threshold alone cannot distinguish between the two confidence patterns of "0.86 vs. 0.10" and "0.86 vs. 0.84"; the stable window condition is used to suppress decision jitter caused by fluctuations in confidence around the threshold, avoiding frequent reversals of semantic conclusions that lead to repeated reprojection of downstream domain models, and is a necessary design to ensure cross-domain synchronization stability. Semantic content that does not meet the convergence condition can legally exist in the hypothesis field as a multi-hypothesis state, without blocking the modeling process or writing it into the determined semantic state.

[0071] Conflict arbitration. When there is a conflict in the support for the same semantic information item between semantic evidence records, and the degree of conflict (preferably measured by the total variance distance between the support distributions of the two pieces of evidence) exceeds a preset conflict threshold. i conf At that time, weighted arbitration is carried out according to the reliability of the source of evidence; conflicts that cannot be automatically resolved generate manual review items and are presented through the access and interaction layer. The manual review results are used as new semantic evidence records to participate in the fusion again. The review results are also not directly written in the state and the rules for evidence entry are followed.

[0072] After the fusion and acceptance are completed, the dynamic semantic state field establishes or updates a formal bidirectional index between the semantic evidence fragments and the accepted semantic conclusions; the initial index between the fragments established during evidence generation and the candidate semantic hypotheses only represents the candidate support relationship, and the two are clearly distinguished.

[0073] (v) Task relative completeness, proactive completion and hierarchical generation gating

[0074] Each target projection d Predeclare the set of semantic information items required to generate the corresponding domain model. s d Each semantic information item v ∈ s d weight w v and determining the threshold i v det Among them, each weight w v >0, to ensure that the denominator of the completeness of the weighted sub-completion of equation (7) and the optional sub-completion described below is strictly greater than zero; s d Further categorized into a subset of mandatory semantic information items. s d req Optional semantic information item subset s d opt A subset of semantic information items associated with hard constraints s d hc Unified semantic projection range V Relative target projection d The definition of subcompleteness is given in equation (7): In equation (7): 1[·] is the indicator function; conv( v () is a semantic item v The convergence indicator is given if and only if v The corresponding mutually exclusive candidate semantic hypothesis group takes the value 1 when it satisfies the convergence criterion of equation (6); b v This represents the confidence level of the highest candidate hypothesis for the current information item. Specifically, completeness only counts converged semantic information items: information items in a multi-hypothesis state are not included even if their highest confidence level reaches a certain threshold. Therefore, completeness becomes an operator defined on the output of the convergence mechanism, rather than a general readiness check of the original data. (Divided by the statistical target projection defined in equation (7)) d The subcompleteness Cmp of all declared semantic information items d ( V In addition, a subset of optional semantic information items is defined using the same calculation method. s d opt Optional weighted subcompleteness Cmp d opt ( V Its molecules are: s d opt The sum of the weights of semantic information terms that satisfy the convergence criterion of equation (6) and have a confidence level not lower than the corresponding determined threshold, with the denominator being: s d opt The sum of the weights of all semantic information items; when s d opt When it is an empty set, Cmp is defined. d opt ( V =1 to avoid the denominator being zero and to ensure that the corresponding gating condition is satisfied. Distinguish between Cmp and other gating conditions. d ( V ) and Cmp d opt ( V The necessity of this is that the required semantic information items and the semantic information items associated with hard constraints have already been determined item by item by independent conditions. If the formally generated conditions are then used with Cmp containing required items... d ( VIf weighted gating is used, the same information item will be counted repeatedly, and the degree of absence of optional items will be masked by the satisfaction of mandatory items.

[0075] Completeness is determined on a per-target projection basis. Only when the set of target projections... G When each target projection in the process is required to be atomized and generated simultaneously, the completeness of each sub-projection is aggregated according to equation (8) to be relative to the total completeness of the target projection. G Set completeness: (8) In equation (8), Agg(·) is a preset aggregation rule, which preferably takes the minimum value of the completeness of each sub-completion. Alternatively, a weighted average can be used. The fundamental rule is object-by-object projection gating: even in cases requiring simultaneous atomization, gating only occurs when... G Simultaneous generation is only performed when each target projection satisfies its own formal generation conditions. The aggregated value in equation (8) is only used as a statistical indicator to present the overall readiness to the user and does not replace the generation gating on a per-target-projection basis. If the aggregated value, especially the weighted average, is used directly as the gating basis, a target projection that does not yet meet the generation conditions may be masked by the average of the other highly complete target projections, thus causing the domain model that does not meet the generation conditions to be generated together. This is the path excluded in this design. The technical significance of distinguishing between per-target gating and aggregated indicators is that: the semantic information items required by different target projections are different from each other, and the missing information items of one target projection may not be necessary for another target projection; if the global minimum value is used for gating, the target projection that has met the generation conditions will be blocked by the missing information items that are irrelevant to it, resulting in unnecessary waiting and invalid completion. The completeness is calculated according to the set of target projections selected in this round of tasks. Different tasks can draw different conclusions, reflecting the relativity of tasks.

[0076] Hierarchical gating is generated. Gating is applied to each target projection. d Execute them separately. Assume the target projection... d The generation threshold is c gen,d The draft threshold is c draft,d And 0 < c draft,d < c gen,d If the value is less than or equal to 1, the gating result of the target projection is denoted as Out. d The gating rules are shown in equation (9): In equation (9): P1 P1 indicates the execution of formal projection, establishment of projection relationship records, and entry into semantic conservation quantity verification; P2 indicates the generation of a draft state domain model, which undergoes structural validity checks and non-exemptible hard constraint checks. If the checks pass, the draft state is displayed via the access and interaction layer; if the checks fail, it is not displayed but a draft generation failure message is output. Simultaneously, active completion can be initiated as described in this section to improve completeness; P3 This indicates that active completion is performed, and the target projection is marked as incomplete and presented after the active completion termination condition is met. The difference between the two lies in whether a draft state domain model is output simultaneously, rather than whether active completion is performed. d Projecting the target d The formal generation condition indicator is 1 if and only if all three of the following conditions are met: (1) s d req All required semantic information items satisfy the convergence criterion of equation (6) and the confidence level is not lower than the corresponding determined threshold. i v det (2) s d hc All hard constraint-related semantic information items in the equation (6) satisfy the convergence criterion and have a confidence level not lower than the corresponding determined threshold, so that the relevant hard constraints have the conditions for evaluation. Among them, the non-exempt hard constraints are satisfied after deterministic evaluation, and the exemptable hard constraints are satisfied after deterministic evaluation, or have been effectively exempted according to the explicitly configured exemption rules and the exemption basis is recorded in the evolution field. Here, "no unresolved conflict" is not used to replace "sufficient information and satisfied constraints", nor is the exemption path allowed to be applied to non-exempt hard constraints. The reason for the former is that the two are not equivalent. When there is no evidence for a certain hard constraint-related information item, Neither conflict nor evaluation is possible. If only the absence of conflict is taken as the condition, the power consumption limit is fixed but the current power consumption is missing, interference constraints exist but the key dimensions have not converged, protocol compatibility constraints exist but the interface type is not determined, etc., can all be gated. Models lacking key security information will enter formal generation based on this. The reason for the latter is that the unexempt hard constraints, according to the definition of hard constraints and soft constraints in the terminology section of this specification, do not have an exemption path. If exemption is also opened for them, this type of constraint will no longer be substantially different from the exemptable hard constraints. (3) Optional weighted subcompleteness Cmp d opt ( V Not less than c gen,d , s d optWhen the set is empty, the item is considered satisfied. The effect of judging the required items and hard constraints independently is to prevent the weighted proportion from masking the lack of key interfaces or security constraints. If only weighted numerical gating is used, the absence of a constraint item with a low weight but necessary for security may still make the completeness value meet the standard. The draft state domain model does not participate in the comparison of semantic conservation quantities and is not eligible for synchronization. Its content itself is not imposed on any other domain model. When the draft state domain model is generated, a draft projection relationship record with a draft state identifier is established at the same time. The user's editing on the draft is located to the corresponding unified semantic object or attribute item by the lifting operator registered in the target domain according to the draft projection relationship record, and a draft change evidence record is generated. The record is decomposed into atomic model changes, generates semantic change candidates and performs the same set of state judgments according to the rules described in Part (VII). The benchmark model version comparison of the pre-validation check is based on the current draft model version and the task version identifier, not the positive Based on the domain model version, the edits on the draft may also contain content that violates hard constraints, semantic ambiguity, or low-reliability candidates. Therefore, it is not allowed to bypass the state splitting and directly enter the fusion: those who have obtained the accepted state participate in the fusion, those who are in the candidate retention state participate in the fusion but their results only update the confidence in the hypothesis field, those who are in the conflict state and the rejection state do not participate in the fusion, and those who are in the pending user confirmation state enter the field with the user's confirmation or modification results as new semantic evidence records; the confidence of the corresponding candidate semantic hypothesis is updated accordingly; among them, those who are in the candidate retention state and those who have obtained the accepted state but have not formed accepted semantic changes after fusion, their corresponding semantic information items are not included in equation (7). The completeness statistics are only calculated after the draft change candidate has been fused to form a recognized semantic change and written into the determined semantic state. Only then does the semantic information item satisfy the inclusion condition of equation (7) and be recalculated accordingly, which may increase the completeness of the target projection. The statistical range of completeness is always limited to semantic information items that have satisfied the convergence criterion of equation (6) and whose confidence is not lower than the corresponding determined threshold. The draft path does not constitute an exception to this rule. The draft change evidence record and the domain change evidence record for the formal domain model adopt the same record structure and are written into the evidence field in the same appended immutable manner. The difference is that its source model does not have the qualification to be synchronized, so it does not involve the revocation and restoration of the source model's synchronization qualification, nor does it perform the semantic conservation quantity closed-loop verification of the source model on the draft model. However, once the edits on the draft are accepted and converged to form accepted semantic changes, the consequences are no different from accepted semantic changes from other sources: the determined semantic state of the unified semantic object has changed, and the formal domain models of each affected target domain have become outdated versions relative to the updated unified semantic projection range. Therefore, it is necessary to generate unified semantic state change events as described in Part (VII), temporarily revoke their synchronization eligibility, and perform incremental reprojection. Otherwise, the dynamic semantic state field will no longer be the single source of facts for all formal domain models.It is important to distinguish between "the content of the draft model" and "accepted semantic changes derived from the draft editing": the former is never applied to other domain models, while the latter, along with semantic changes caused by positive evidence input, is subject to the same set of gating and the same propagation path. What truly needs to be isolated is the unverified draft product, not the semantic conclusions that have been verified through convergence criteria and hard constraints. (To be Generated) d After setting 1, the formal projection is re-executed from the updated unified semantic projection range and verified by conservation quantities. The verification path does not transfer draft content, ensuring that the object of conservation quantity verification is always the product in a complete state. After verification, the original draft model and its draft projection relationship record are marked as replaced or archived, and draft change events are no longer accepted. Continued editing requires creating a new draft branch and a new draft projection relationship record based on the current formal domain model or the current unified semantic projection range. When there are changes in the version of the task involving the projection operator version, semantic information item structure, or target projection configuration, the original draft model and its draft projection relationship record are marked as pending reconstruction. The draft projection is re-executed according to the new task version. Draft modifications that have not yet been submitted by the user are improved after performing a three-way comparison or rebasing on the regenerated draft model. Furthermore, the domain models corresponding to each target projection in the target projection set can be generated either all at once or in stages: if a target projection is delayed because its declared semantic information items have not yet converged, a target projection that has met the generation conditions can be formally projected first and obtain synchronization eligibility after being verified by the conservation quantity; after the remaining target projections meet the generation conditions, they will be projected, and their products, after passing the verification, will also obtain synchronization eligibility and participate in cross-domain change propagation. Staged generation does not change the constraint that cross-domain change propagation is executed through the unified semantic projection scope.

[0077] Proactively address and constrain violations. When Gen d When the value is 0, it is necessary to first distinguish whether the cause is insufficient information or a violation of constraints, as the handling paths for the two are completely different. For those caused by insufficient information (the existence of semantic information items that have not converged, have not reached a certain threshold, or have unresolved conflicts, or the relevant hard constraints do not yet have the conditions for evaluation due to the lack of associated information items), the completeness determination module locates the semantic information items to be completed under the target projection and places them in the candidate collection action set Act. Select the data collection action according to formula (10). This applies to cases where the cause is determined to be a constraint violation. s d hcAll hard constraint-related semantic information terms have converged and reached the confidence level. The hard constraints have met the evaluation conditions, but deterministic evaluation shows that they are not satisfied. For example, the estimated power consumption has converged to 38 W, while the upper limit is clearly 35 W. W, at this point, the set of semantic information items to be completed is empty. Continuing to ask the user questions, schedule sensors, or call external artificial intelligence models will not change the evaluation result. Therefore, no data collection action is performed, and the following path is followed: When the evaluation of non-exempt hard constraints is not satisfied, the generation of the formal domain model for the target projection in this round is stopped. The output indicates the violated constraints, the basis for their evaluation, and the constraint violation prompt information of the involved semantic information items, which is presented through the access and interaction layer. Correction suggestions for the values ​​of the relevant semantic information items can also be generated. This correction suggestion is calculated by the system based on the existing semantic state and constraints. It belongs to the candidate solution rather than the new independent observation of the real object. Therefore, it is only output as a correction candidate or interaction suggestion and is not directly used as semantic evidence for fusion. Only after the user selects, modifies, or confirms the correction suggestion, or after the correction suggestion is verified by independent constraint solving, simulation, or deterministic verification, or after external files, sensors, or other information sources provide new input data that supports the value of the suggestion, will a new semantic evidence record be formed for fusion. If the automatically generated correction suggestions are allowed to enter the field, a self-verifying loop will be formed: "the system calculates suggestions based on the current situation—converts suggestions into evidence—and then uses the evidence to support the value of the suggestions." This contradicts the discipline of "decoupling generation and acceptance" in Part (II) and the general prohibition of single-source self-verification in Parts (VIII) and (X). Without any of the above circumstances, the correction suggestion does not form a semantic evidence record, but is reserved as a correction suggestion candidate in the restricted system candidates defined in the terminology definition section of this specification. It belongs to the same restricted system candidates as the historical prediction candidates generated according to the evolutionary history: it can participate in the confidence calculation of the corresponding candidate semantic hypothesis according to the restricted weight of Equation (14), but it cannot be used as the sole source category to make the corresponding semantic information item obtain a determined semantic state. Here, a single rule is used instead of the two treatments of "not allowed to enter" and "low weight entry" because the two give mutually exclusive conclusions for the same situation, which will make the evidence entry discipline uncertain. The rule of this invention is: the correction suggestion does not have the identity of a semantic evidence record before it is verified, and its upper limit of influence is the restricted weight of the historical prediction candidate. If the evaluation of an exemptable hard constraint is not met and a valid exemption has not yet been obtained, the process proceeds to either a value correction process or an exemption approval process based on explicitly configured exemption rules. Without this distinction, the final state will be suspended in the case of "generation conditions not met, all items to be supplemented are empty, and no data collection action can be selected," which contradicts the requirement of "all final states are observable" described in Part (XII) of this specification. Data collection actions for reasons of insufficient information are selected according to formula (10): (10) In formula (10): oTo perform the data collection action act Expected observations; Unc( V (This refers to the weighting of the semantic information items to be completed) w v The weighted current uncertainty is preferably measured by the weighted summation of the Shannon entropy of the distribution of candidate hypotheses for each missing information item; Unc( V | o To obtain observations o The conditional uncertainty following; 𝔼 o~act [·] This represents the expected value of the observed distribution; the numerator is the expected reduction in uncertainty (expected information gain); cost( act The cost of the data acquisition action is denoted as . This formula is the engineering form of the expected information gain criterion in Bayesian experimental design, enabling the system to "know what it lacks and obtain it in the most cost-effective way." Data acquisition actions include asking the user questions, requesting supplementary documents, scheduling sensors to re-acquire data, and calling external artificial intelligence models. The supplementary input data returned is then used by the evidence generation module to regenerate semantic evidence records, which enter the fusion loop. The current round of completion stops when preset active completion termination conditions are met (such as reaching the maximum number of acquisitions, no available acquisition actions, or user rejection). The unified semantic projection range is marked as incomplete relative to the target projection, and an incomplete prompt indicating the semantic information items to be completed is output and presented through the access and interaction layer.

[0078] (vi) Semantic conservation quantities and consistency verification gating.

[0079] Define a shared conserved quantity space CQS, whose elements are normalized semantic conserved quantity vectors CQ, and whose components include at least the object identity identifier CQ. id and topological relationship structure CQ top Key parameter CQ normalized to units of measurement par Standardized Interface Signature (CQ) int Standardized effective constraint set CQ con At least one of the following; each target projection declares its set of required components, set of optional components, deviation threshold for each component, and normalization rule version. The absence of components not declared as required does not constitute a verification failure. The baseline semantic conservation quantity CQ( V )∈CQS is calculated directly from the unified semantic projection range; each target domain d Conservative quantity extraction logic Ext d Its domain model M d = P d ( VMapping back to CQS. Conservative quantity extraction uses the generated domain model as input and is re-extracted via the domain's model parsing path, without reusing intermediate results generated by the projection operator during the generation of the domain model. The extraction logic and projection operator can share the domain meta-model and normalization base library, but the projection operator version, conservative quantity extraction configuration version, and normalization rule version are recorded separately and stored in the evolution field along with the verification results. The necessity of this design lies in the fact that if conservative quantities are directly taken from the intermediate projection results, errors in the projection rules will be simultaneously reflected on the baseline and extraction sides, canceling each other out, inevitably leading to a common-mode error. Re-parsing with the generated model as input ensures that the verification path and the generation path are independent in the input and parsing stages, thus giving the verification practical discriminative power. Furthermore, the object identity set of the baseline semantic conservation quantity only includes object identities that have satisfied the convergence criterion of equation (6): Unified semantic objects whose object identities have not yet converged can be included in the unified semantic projection range to carry relational objects and constraint objects on them, but they do not participate in the current round of projection generation, and their identity identifiers are not included in the baseline semantic conservation quantity. Therefore, there are no corresponding model elements in each domain model that constitute the identity identifier component deviation. After their object identities satisfy the convergence criterion, the object is included in the corresponding domain model through influence range determination and incremental reprojection, and is included in the object identity set of the baseline semantic conservation quantity from this round onwards. Similarly, relational objects whose object identities have not yet satisfied the convergence criterion will not participate in the projection generation in this round, and their identity identifiers and topology are not included in the baseline semantic conservation quantity. After the object identities of the object at one end converge, they will be included together with the object and supplemented into the corresponding domain model through incremental reprojection. This rule ensures that unconverged objects neither block target projections that have the conditions for generation nor cause false failures in verification due to the lack of corresponding model elements.

[0080] Consistency deviation is calculated by components. Let the first component be... r The normalization bias of the conserved component is d d,rThe measurement method is standardized according to component type: the comparison of identity components is performed in the standardized identity space. The local identifiers of model elements in each domain (such as SysML element identifiers, CAD feature identifiers, and UML component identifiers, whose namespaces and encoding rules are usually different) are first restored to the standardized identifiers of the corresponding unified semantic objects according to the identifier mapping rules and projection relationships of the target domain. Then, they are compared with the standardized identifier set in the baseline semantic conservation quantity by exact matching, and the single deviation is taken as 0 or 1. Among them, the model elements that cannot be restored to any standardized identifier, the standardized identifiers that exist in the baseline but do not have corresponding elements in the model of this domain, and the cases where the same standardized identifier corresponds to multiple model elements that should be unique are all counted as identity component deviation. The discriminative power of identity components comes from the missing, redundant, and disordered correspondence of model elements, rather than the literal consistency of local identifier strings, thus adapting to the actual situation that the identifier systems of model elements in different domains are different. The topological relationship structure preferably adopts the normalized graph edit distance; the key parameters adopt the normalized relative deviation under the baseline dimension; the interface signature and constraint set adopt the normalized set symmetric difference ratio. The total consistency deviation is shown in equation (11): (11) In equation (11): l d,r ≥0 represents the target domain d For the r Deviation weights of conserved components; i d cons The overall consistency deviation threshold for this area is set, and separate thresholds are set for each component. i d,r cons If a single deviation or the total deviation exceeds the corresponding threshold, the semantic consistency condition is deemed not met. Only domain models that meet the condition are marked as synchronizable formal domain models; the verification is performed on a domain-by-domain basis, and the failure of one model does not block other models.

[0081] As a preferred mathematical extension, multi-domain consistency can be uniformly expressed as a consistency problem on the cellular sheaf: a layer structure is constructed with each target domain and the unified semantic projection range as nodes, and the extraction logic of conserved quantities and the projection relationship as constraints, and all conserved quantities are assigned values. x The global inconsistency is measured by the quadratic form of the layer Lap matrix, as shown in equation (12): (12) In equation (12): x The assigned vector is formed by concatenating the conserved quantities extracted from each domain with the baseline conserved quantities; Lap is the layer Laplacian matrix constructed according to the constraint mapping. i glob This is the global consistency threshold; Inc( x =0 if and only if the assignment is a global section (all regions are completely consistent), the component decomposition of the quadratic form can directly locate the region where the deviation occurs and the conserved component. This extension is an optional implementation. The component comparison of equation (11) constitutes a complete implementation scheme. Those skilled in the art can also use other equivalent consistency measurement methods to achieve the same function.

[0082] When the verification is not satisfied, the repair loop is entered: the deviation component is located, a repair proposal is generated, and a re-comparison is performed after the repair is executed; the repair proposal is divided into three categories according to the repair object, and different execution disciplines apply to each: the repair proposal for the semantic content within the scope of the unified semantic projection does not directly rewrite the unified semantic object; and since it is calculated by the system based on the existing semantic state and the measured deviation, and is not a new independent observation, it is not directly used as semantic evidence: only after being selected or confirmed by the user, verified by independent constraint solving or deterministic verification, or after obtaining independent source input data other than the repair proposal, is it submitted for fusion in the form of a new semantic evidence record; if the above situations do not occur, the repair proposal does not form a semantic evidence record, but is reserved as a repair proposal candidate in the restricted system candidate defined in the terminology definition section of this specification, and belongs to the restricted system candidate along with the historical prediction candidate generated according to the evolutionary history, and can be calculated according to formula (14). The restricted weights of the data are used in the confidence calculation of candidate semantic hypotheses, but they cannot be used as the sole source category to make the corresponding semantic information item obtain a determined semantic state, so as to maintain the discipline of evidence entry and avoid self-confirmation of the repair loop; for repair proposals for projection operators or configuration rules of projection operators, after forming a new operator version or configuration version, the projection and consistency verification module will re-execute the projection based on the updated operator; for repair proposals for domain models, they are limited to repairs that do not change the serialization format, reference integrity and naming conventions of the model's semantic content. They cannot pass the verification by directly rewriting the semantic content of the domain model. If the root cause of the semantic content deviation lies in the unified semantic projection range or projection operator, directly rewriting the domain model can only temporarily eliminate the deviation, and the deviation will still be reproduced in the next projection. When the preset repair termination conditions are met (such as reaching the maximum number of repairs, no available repair proposals), the repair will stop, the corresponding domain model will be marked as a consistency verification failure state, will not be included in the automatic cross-domain synchronization link, and a verification failure prompt message containing the located deviation component will be output and presented by the access and interaction layer. Without this gating, the inconsistency of propagation followed by discovery would cause errors to spread across multiple domains and make it difficult to pinpoint the source. This gating ensures that every domain model has passed the semantic fidelity test before being eligible for synchronization, without exception for the initial projection product and the incremental reprojection product. Draft-state domain models do not participate in the comparison of semantically conserved quantities, nor do they obtain synchronization eligibility. They are only presented as draft status after passing structural validity checks and non-exemptible hard constraint checks through the access and interaction layers. If the checks fail, they are not presented, but a draft generation failure message is output. Therefore, the "passing the semantic fidelity test" mentioned in this specification is limited to domain models that have obtained synchronization eligibility and does not cover draft-state domain models.After the repair of the projection operator or its configuration rules is executed, a new operator version or configuration version is formed. The projection relationship records that depend on the old version are marked as to be rebuilt and re-established when the projection is re-executed according to the new version. The domain model obtained by re-projection must be re-executed to compare the semantic conservation quantity of Equation (11) before it can be synchronized. Other target domain projection relationship records and domain models that do not depend on this version are not changed as a result.

[0083] (vii) Changes in evidence return, scope of impact and incremental reprojection.

[0084] When the first formal domain model that can be synchronized in the first target domain changes (user editing, external tool feedback, or requirement document update; the requirement document itself can also be a domain model), the system does not directly apply the change to other domain models, but performs evidence-based backflow: generating an append-only immutable domain change evidence record (including change source, operating entity, change time, model difference before and after the change, projection instance handle, baseline semantic version identifier, baseline model version identifier, and idempotency identifier), and the lift operator registered in the first target domain. L dThe change is handled using a promotion mode. When a change event occurs, the synchronization eligibility of the first formal domain model after the change is temporarily revoked, preventing it from participating in cross-domain change propagation before the completion of this change evaluation and closed-loop verification. The content of the changed model has not yet been verified by semantic conservation. If it still retains synchronization eligibility during the evaluation period, the unaccepted edited content may spread through other links. The promotion operator first decomposes the model difference into one or more atomic model changes based on the projection relationship record. Each atomic model change generates semantic change candidates involving one or more unified semantic objects, relation objects, or constraint objects in the unified semantic projection range. The reason why atomic model changes can be located to specific relation objects or constraint objects is that all relations and constraints that generate independent model elements (such as connectors, dependency edges, constraint blocks, and their binding relationships) in the target domain have been materialized into relational or constraint unified semantic objects as described in Part (I), thus having independent object identifiers and serving as mapping units for projection relationship records. Inline relations and inline constraints that have not been materialized do not generate independent model elements separately. Their manifestation in the target domain is projected along with their respective unified semantic objects. If unintegrated inline items are allowed to directly generate independent model elements, then projection relationship records can only be recorded up to their parent object. When the same parent object holds multiple inline relationships, it becomes impossible to distinguish them. Backward promotion will not be able to locate specific relationships or constraints, and the scope of influence will be forced to expand to the entire parent object, thus degrading incremental reprojection. Since a single edit often modifies entities, relationships, and constraints simultaneously (such as adding a new bus while changing the port topology, interface list, and power consumption constraint satisfaction status), if processed on a per-difference basis, it will be impossible to distinguish which part is accepted and which part is rejected.After each semantic change candidate is generated, it is placed in an evaluation state and categorized according to the following priority based on the reliability of the source and the constraint verification results: Candidates that violate non-exempt hard constraints as confirmed by deterministic evaluation are placed in a rejected state; candidates that cannot be determined whether they violate non-exempt hard constraints because the candidate value itself has not yet converged, the scope of application of the constraint is not determined, or there are unresolved conflicts in the supporting evidence are placed in a conflict state; Candidates that violate exemptable hard constraints as confirmed by deterministic evaluation but have not yet completed a valid exemption according to the explicit exemption rules are also placed in a conflict state. After completing a valid exemption, the state determination is re-executed. If the exemption application is rejected or the preset exemption processing conditions are exceeded, it will be transferred to a rejected state. The reason why such candidates are not directly placed in a rejected state is that exemptable hard constraints have a legitimate exemption path according to the definition of hard constraints and soft constraints in the terminology definition section of this specification. Before the exemption procedure is completed, its acceptability is not yet determined, which is different in nature from the certain violation of non-exempt hard constraints; Candidates that do not meet the acceptance criteria are not fused or updated before resolution. Candidates that do not violate hard constraints, do not have semantic ambiguities that cannot be automatically resolved, and meet the acceptance criteria are set to the acceptance state. After fusion by equation (5), if the fusion result meets the convergence criterion of equation (6) and there are no unresolved hard constraint conflicts, it constitutes an accepted semantic change and updates the determined semantic state of the corresponding object. If the convergence criterion is not met after fusion, only the confidence of the corresponding candidate semantic hypothesis in the hypothesis field is updated. Candidates with semantic ambiguities that cannot be automatically resolved are set to the pending user confirmation state. After the access and interaction layer outputs a confirmation request, the user confirmation or modification result is used as a new high-reliability semantic evidence to enter the fusion, but is not exempt from hard constraints. Other candidates that do not meet the acceptance criteria are set to the candidate retention state. Such candidates still participate in the fusion of equation (5), but their fusion result only updates the confidence of the corresponding candidate semantic hypothesis in the hypothesis field and is not written into the determined semantic state. Thus, they can be overwhelmed or elevated by subsequent evidence without re-execution of the elevation. The technical significance of distinguishing between rejection and conflict states lies in the completely different subsequent paths of the two types of candidates: a rejection state indicates that the candidate has been determined to be unacceptable, and its corresponding model element, relationship, or attribute should immediately undergo local rollback or isolation marking; a conflict state indicates that there is still a possibility of acceptance after supplementary evidence or constraint resolution, and the candidate should be retained and awaited. Taking hard power constraints as an example, if the estimated power consumption after modification is deterministically determined to be 38 W while the upper limit is clearly 35 W, then the corresponding candidate is placed in a rejection state; if the power consumption parameter itself still has multiple conflicting pieces of evidence, and it is impossible to determine whether the limit is exceeded after modification, then the corresponding candidate is placed in a conflict state; without this distinction, state routing will be arbitrary, and the same situation may lead to different treatments in different implementations, undermining the determinism of gating.The reason for placing pending user confirmation before candidate retention is that the initiator of the change event is usually the current user, and any ambiguity regarding their editing intent should be verified immediately rather than left for later evidence. When the two conditions overlap, prioritizing verification can significantly shorten the duration of ambiguity. Candidate retention is a fallback state when the strength of evidence is insufficient. When only a portion of the semantic change candidates for the same change event are fused to form accepted semantic changes, the unified semantic object is updated only based on the accepted semantic changes and a verification expectation model is generated accordingly. For candidates that have not obtained acceptance status, and for candidates that have obtained acceptance status but have not yet formed accepted semantic changes after fusion, the corresponding model elements, relationships, or attribute items are partially rolled back, isolated, or user confirmation requests are output. Before all the above candidates are processed, the changed model is not restored to synchronized status as a whole, so as to avoid the edited content that has not been accepted by the semantic base flowing into other domains along with the synchronized status of the model. When the projection instance handle is valid, the object is directly located according to the projection relationship record. When the handle is missing, invalid, or ambiguous, the evidence object is anchored according to formula (4). If the baseline semantic version identifier recorded in the domain change evidence record is inconsistent with the current semantic version, the promotion and conflict determination will be re-executed on the current semantic version to avoid concurrent editing from overwriting each other.

[0085] The conversion from atomic model changes to semantic change candidates follows these rules, depending on the type of model change: When adding a new model element, the corresponding semantic element category is retrieved based on the element's type in the domain metamodel, generating semantic change candidates such as "adding a unified semantic object," "adding a relational object," or "adding a constraint object," with the location information taken from the element's path identifier in the model; When deleting a model element, the corresponding semantic element is retrieved based on the projection dependency index, generating two mutually exclusive candidates: "set the semantic element to a tombstone state" and "hide the semantic element only in the domain model." These two candidates have different semantic consequences and are usually not distinguishable by the model difference itself. Therefore, based on the priority of traffic allocation, the status is set to await user confirmation. When updating the attribute values ​​of model elements, candidate semantic hypotheses for attribute values ​​corresponding to a unified semantic object or attribute item are generated. Their initial evidence support is set according to the reliability of the change source and, after normalization by measurement unit, they enter the same mutually exclusive candidate group as existing hypotheses. When reconnecting the relationships between model elements, two semantic change candidates, "remove the original relation object" and "add a new relation object," are generated. Both are evaluated together as related candidates for the same atomic model change. When the change involves model content expressed in the form of constraints, candidate expression changes for the constraint objects are generated, and hard constraint verification is performed before acceptance. When the same atomic model change can be located to multiple semantic elements based on the projection dependency index, semantic change candidates are generated for each semantic element and evaluated independently. The rejection of any candidate does not affect the acceptance of the remaining candidates. For semantic change candidates that have been accepted through evidence fusion, before entering the scope of influence determination and source model closed-loop verification, the formal projection relationship records on the source formal domain model side are processed according to the atomic model change type corresponding to the semantic change candidate: when adding a new model element, adding a new association, or adding a new constraint, a projection instance handle is generated for the source model element and a formal projection relationship record is established based on the unified semantic object identifier obtained by association matching or evidence object anchoring; when updating a model element or attribute item, the source semantic version, target model version, and attribute item identifier in the existing formal projection relationship record are updated; when reconnecting associations, the endpoint reference of the relational unified semantic object and its formal projection relationship record with the associated elements of the source model are updated; when a model element, association, or constraint is deleted, the corresponding formal projection relationship record is marked as terminated or invalid, depending on whether the accepted semantic change indicates semantic deletion or only the removal of the domain projection product. All of the above processes retain historical records by adding projection relationship change events. After successful processing, the corresponding model content is entered into the current valid semantic view of the source formal domain model as the current valid projection instance. If the processing fails, the corresponding projection relationship record is marked as pending reconstruction, the source formal domain model's synchronized qualification is revoked, and the formed accepted semantic changes are not rolled back. The qualification will be restored after the projection relationship record is reconstructed and the closed-loop verification is passed.For semantic change candidates that have not been accepted as semantic changes, no new formal projection relationship record will be established or activated, and the corresponding model content will be partially rolled back or isolated according to the candidate state.

[0086] The propagation phase only begins when the semantic change candidates satisfy the convergence criterion of equation (6) after fusion and there are no unresolved hard constraint conflicts: the impact analysis and incremental reprojection module determines the scope of influence based on the relationships and constraints in the updated unified semantic projection range, as shown in equation (13): (13) In formula (13): ΔV This is the set of changes to the unified semantic projection range caused by this fusion; Cl( ΔV ) represents the set of semantic elements covered by the closure reachable from the relation field and constraint field of this change set; Prj V To unify the set of projection instances contained in the projection relationships of objects within the semantic projection scope, p For any of these projection instances; or For any semantic element; reach( or , p ) is an reachability indicator function if and only if the semantic element or Projected instance p The reference value is 1, and its reference relationship is given by the projection relationship record. Projections outside the scope of influence are not touched, achieving incremental rather than full synchronization. The reachable closure Cl( in equation (13) ΔV ) Calculate according to the following propagation configuration to ensure that propagation is bounded and that the "increment" property holds: (1) Predeclare the set of propagable relation types. R prop only R prop Relation types in the set participate in propagation, while relation types not in the set (such as relations used only for grouping, namespace affiliation, etc., which do not carry semantic dependencies) do not participate in propagation; (2) for R prop Each relation type in the relation declares the propagation direction, taking one of forward, reverse, or bidirectional. For example, the parameters that make up the relation from the component to the whole are considered forward propagation, and the constraints from the whole to the component are considered reverse propagation; (3) declare the maximum propagation depth. h max ,since ΔV Each semantic element in the text is traversed using either breadth-first or depth-first search, exceeding... h max (4) Maintain a set of visited semantic elements. The same semantic element is visited at most once in a closure calculation. Based on this, the calculation can still terminate when there is a loop in the relation graph. (5) Constraint dependencies are distinguished by their type: semantic elements referenced by non-exempt hard constraints are forcibly included in the closure and are not subject to the constraints. h max Constraints are imposed to ensure that the satisfied state of hard constraints is always recalculated; the semantic elements referenced by soft constraints are... R prop and h max The general rules of propagation; (6) When there are no new propagable relations to expand, the maximum propagation depth is reached, or the projection boundary given by the projection dependency index is hit during traversal, the current closure calculation is terminated. If the above configuration is missing, a local change in the relation graph may propagate infinitely along any relation to all objects, causing incremental reprojection to degenerate into full reprojection; the above configuration makes the scale of the scope of influence determined by the declarative configuration rather than by the connectivity of the relation graph. It should be clarified that the semantic changes that trigger the determination of the scope of influence and incremental reprojection are not limited to those generated by the domain model change event through the lifting operator. When semantic input data from any information source, including user-added text or speech, newly arrived sensor observations, newly imported files, new parsing results from external artificial intelligence models, and manual review results, are fused by Equation (5) and satisfy the convergence criterion of Equation (6) without any unresolved hard constraint conflicts, thereby causing changes in the determined semantic content within the unified semantic projection range, the system also generates a unified semantic state change event and determines the scope of influence according to Equation (13): for each target projection affected within the scope of influence, the synchronization qualification of its corresponding formal domain model is temporarily revoked, and then the projection and consistency verification module performs incremental reprojection, and the synchronization qualification of each model is restored only after the obtained domain model passes the semantic conservation quantity comparison of Equation (11). This branch shares the same influence scope calculation logic, the same projection execution point, and the same conservation quantity gating with the propagation caused by the domain model change event. The only difference is that it does not have a "source domain model," and therefore does not perform the semantic conservation quantity closed-loop verification of the source model described in Part (VII). The purpose of this closed-loop verification is to verify whether the user's edit on a certain domain model has been faithfully absorbed by the semantic base. The semantic change does not originate from the absence of the verified object when the domain model was edited. The resulting model update is also marked as a system source update, and no new domain change evidence record is generated for it. If this branch is missing, other domain models are only updated when the domain model is edited. Although a newly arrived demand text or sensor observation has changed the semantic base, each existing formal domain model will still maintain its content based on the old semantic version and continue to participate in cross-domain synchronization, which contradicts the position of this invention that the semantic base is the single source of fact. This branch ensures that any change in the defined semantic content of a unified semantic object, regardless of whether it originates from positive evidence input or reverse model editing, is consistently propagated to each target domain through the same set of gating; unified semantic objects whose identities have not yet converged as described in Part (VI) are incorporated into the corresponding domain model after their identities converge, and structural changes caused by the objectification of attribute items as described in Part (VIII) are also performed in this branch.

[0087] The impact analysis and incremental reprojection module sends an incremental reprojection request to the projection and consistency verification module, which then performs incremental reprojection using the projection operators registered in the affected target domain and performs semantic conservation comparison on the resulting model again. The closed-loop verification of the source domain is independent of the existence of an affected second target domain projection and is performed after all semantic change candidates related to this change have been processed accordingly: when this change results in an accepted semantic change, the expected model or expected semantic conservation quantity for the first target domain is generated based on the unified semantic projection range updated by the accepted semantic change; when none of the semantic change candidates for this change have resulted in accepted semantic changes, and their corresponding model elements, relationships, or attribute items have undergone partial rollback or isolation marking, the expected model or expected semantic conservation quantity for the first target domain is generated based on the current unified semantic projection range that maintains the semantic content determined before the change event; in both cases, the current valid semantic view of the first formal domain model after excluding model elements, relationships, and attribute items that have undergone isolation marking is used as the verification object for the semantic conservation quantity closed-loop verification. Therefore, even if all changes are not accepted, the system can still perform closed-loop verification on the currently valid semantic view after rollback or isolation. This allows the source formal domain model to regain its synchronization eligibility after the verification passes, without lacking closed-loop verification results due to the absence of accepted semantic changes. When there is an affected second target domain projection, the source domain closed-loop verification and the incremental reprojection of each second target domain can be executed in parallel or sequentially. If the source domain closed-loop verification is appended after the incremental reprojection of the second target domain, then when a change only involves semantic elements unique to the domain projection and there are no other target domain projections within the affected area, the closed-loop verification will not be triggered. The changed model will remain permanently in a state of revoked synchronization eligibility due to the lack of verification, resulting in a deadlock. This design eliminates this path. It is important to clarify that the expected model or expected semantic conservation quantity used for verification is only an internal verification product of the system. The closed-loop verification does not overwrite the modified first formal domain model with the generated result, nor does it output the verification product as a system source update or model change event. Its purpose is to verify whether the user's changes have been faithfully absorbed by the semantic base, or to verify whether unaccepted model edits have been rolled back or isolated from the current valid semantic view, rather than writing back the source model. When the deviation is within the threshold, the modified first formal domain model regains its synchronization eligibility; when the deviation exceeds the threshold, a repair proposal is generated or a manual confirmation request is output according to the repair loop in Part (VI). Before the handling is completed, the model does not participate in subsequent cross-domain propagation. If direct write-back is used here, the user's newly completed edit may be overwritten by the system-generated result, resulting in the loss of modification and interference with the echo suppression mechanism.All projection executions converge to a single execution point—the projection and consistency verification module—making semantic conservation verification a structural prerequisite for obtaining synchronization eligibility rather than a coding convention. All domain models seeking synchronization eligibility, including initial projection products and incremental reprojection products, must pass conservation verification at this single execution point; there is no path to bypass this verification and enter the cross-domain synchronization link. Draft domain models, although generated from the same execution point, do not participate in conservation comparisons or obtain synchronization eligibility; they are only presented as drafts after passing structural validity checks and non-exempt hard constraint checks. Furthermore, after determining the affected second target domain projection, the synchronization eligibility of the corresponding formal domain model in that second target domain is temporarily revoked: the semantic benchmark upon which the model is based has changed due to the adoption of semantic changes. Before the incremental model patch application and semantic conservation comparison are completed, its content is outdated relative to the updated unified semantic projection range. If it still retains synchronization eligibility, the outdated content will continue to flow to other domains via the synchronization link; the model only regains synchronization eligibility after the incremental model patch is successfully applied and its semantic conservation comparison passes. This revocation is independent of and applies simultaneously to the revocation of the source model's eligibility for synchronization. Together, they ensure that during the handling of a change, there is no domain model that still has eligibility for synchronization despite a change in semantic benchmark.

[0088] Incremental reprojection applies to existing domain models as model patches, rather than regenerating the entire domain model. The projection and consistency verification module, based on the projection dependency index, parses the semantic elements and attribute items falling within the affected scope into a sequence of patch operations acting on specific model elements. These patch operations include at least adding model elements, deleting model elements, updating attribute values ​​of model elements, and reconnecting relationships between model elements. Specifically, adding semantic elements generates a add operation; setting semantic elements to a tombstone state generates a delete operation; changes in attribute function values ​​generate attribute update operations; changes in relational structures generate relationship reconnection operations; and changes in constraint sets generate constraint expression update operations or constrained attribute update operations. Model elements not falling within the affected scope are not rebuilt, and their identifiers remain unchanged. Non-semantic attributes on model elements not determined by semantic elements, including user-adjusted layout coordinates, primitive styles, view collapse states, and annotations, are retained as is during patch execution, only becoming invalid when the corresponding model element is removed by a delete operation. For each affected target domain, the application of model patches, the updating of corresponding projection relationship records, and the comparison of semantic conservation quantities constitute a target local logical transaction. The patch operation sequence is executed in the working copy, shadow version, or transaction-controlled model version in the order of adding elements, updating attributes, reconnecting associations, and deleting elements. When all operations succeed and the semantic conservation quantity comparison of the target domain model passes, the target local transaction is ready to commit. If any operation fails, or the semantic conservation quantity comparison fails and the repair loop has reached the termination condition, the target local transaction is rolled back to the model version and projection relationship record version before the patch execution. The domain model does not regain its synchronizability before this incremental reprojection, but remains in an isolated state or a consistency check failure state and outputs a failure message until the incremental reprojection is re-executed and the semantic conservation quantity comparison passes. This invention supports the following two transaction commit modes: the first is the target independent commit mode. In this mode, target local transactions in each affected target domain are prepared, verified, and committed separately. If a model patch application fails or the semantic conservation comparison fails, only the domain model and its projection relationship records for that target domain are rolled back. Other target domain models that have been successfully committed and passed the semantic conservation comparison are not stripped of their restored synchronizeability due to the failure. Failed targets remain in an isolated state or a consistency check failure state and do not participate in the automatic cross-domain synchronization link, while successfully committed target domain models remain consistent with the current unified semantic state version and can continue to participate in synchronization. This mode is used in the scenario described in claims 22 and 23 where the synchronizeability of each second target domain is restored separately. The second type is the cross-target atomic transaction mode.In this mode, all target domains triggered by the same unified semantic state version and included in the same transaction participant set first execute the preparation phase of the target local transaction separately; only after all participating targets have successfully completed patch operations, passed structural validity checks, and passed their respective semantic conservation comparisons, are the model version pointers and projection relationship record versions of each target domain uniformly submitted, so that each domain model can simultaneously see its new version externally; if any participating target fails to prepare, patch application fails, or semantic conservation comparison fails, compensation operations are performed on all participating targets involved in this transaction or the version snapshot before patch execution is restored. This mode is adopted in the case described in claim 13 where multiple concurrent change events jointly affect the same unified semantic projection range and incremental reprojection is uniformly performed after fusion; this mode is also adopted when the target projection set is configured to require atomic synchronization. The transaction commit mode, transaction identifier, unified semantic state version identifier, identifiers of each participating target domain, preparation status, verification status, commit status, rollback status, and system source update identifier status of each target are all recorded in the evolution field and associated with the version identifier of this round of tasks. Once the same transaction begins execution, it does not switch between target-independent commit mode and cross-target atomic transaction mode. Only valid system source update identifiers are retained for model updates that are ultimately successfully committed; system source update identifiers corresponding to model updates that are rolled back due to transaction rollback are marked as invalid and are not used to suppress subsequent real external change events. Regardless of the transaction commit mode used, the evidence fusion results and unified semantic object versions that have already been written will not be rolled back due to domain model-side transaction failures. Failed targets are kept isolated to prevent the continued propagation of expired models. The above logical transactions do not require the modeling tools of each target domain to natively support the same database transaction. For tools that support shadow versions or working copies, patches are applied to the shadow version and verified before switching the version pointer. For external modeling tools that do not support shadow versions, compensation can be achieved by exporting a copy, applying patches and verifying it, and replacing the file or writing it in batches via the tool interface after verification. The compensation operation involves writing back the copy before execution. Thus, incremental reprojection is semantically equivalent to the minimum necessary modification to the affected model elements, and its execution cost is related to the scale of the affected scope but not to the total scale of the domain model.

[0089] Echo suppression. Model updates actually applied to the domain model by the computer modeling system through incremental reprojection are marked as system source updates. The system generates a system source update identifier for each system source update and stores the system source update identifier in the evolution field in association with the corresponding unified semantic state version identifier, the updated target model version identifier, and the update operation record; a valid system source update identifier is determined only when the system source update identifier and its associated information carried by the model update event are consistent with the internal record in the evolution field. When a model update event carrying a valid system source update identifier is received, no domain change evidence record is generated for the model update event, and no boosting operator processing and evidence fusion are performed; system source update identifiers carried by external inputs but which cannot be matched with the internal record in the evolution field are not determined as valid system source update identifiers and are still processed as ordinary change events. The expected model or expected semantic conservation quantity used for closed-loop verification of the source domain model semantic conservation quantity is only generated internally by the computer modeling system and used for comparison. It is not written into the corresponding formal domain model, does not enter the domain model change event listening channel, and is not considered a system source update. Without the aforementioned system source update identifier and its validity verification mechanism, subtle differences between the incremental reprojection result and the user-edited version may be misjudged as new change events, forming a repetitive cycle of "change—reflow—reprojection—rechange"; this design limits the re-entry point of the loop in step S5 to change events that do not carry a valid system source update identifier, thereby making the loop bounded.

[0090] Concurrent Change Merging. When multiple change events arrive concurrently, the system first obtains or generates an idempotent identifier for each change event according to the rules regarding domain change evidence records, idempotent event indexes, and system source updates in the terminology definition section of this specification, and performs idempotency checks on each event. When a change event carries a source event identifier that is uniquely assigned by its information source within the corresponding source namespace and is not reused, the idempotent identifier is determined by at least the information source identifier, the target domain identifier, the formal domain model identifier, and the source event identifier. When a change event does not carry a source event identifier that meets the aforementioned conditions, the idempotent identifier is determined by at least the information source identifier, the target domain identifier, the formal domain model identifier, the change operation type, the set of identifiers for affected model elements or relationships, the baseline model version identifier, and the normalized change content summary. The operation subject, the information source identifier, a single change object identifier, or a single change content summary are not considered as complete idempotent identifiers. Change events with the same idempotency identifier already existing in the idempotency event index are identified as duplicate change events. Their original received information and the duplication relationship with the original change event are recorded in an append-only, immutable manner. No new domain change evidence records are generated, and the current synchronizable eligibility status of the corresponding formal domain model is not altered due to the duplicate change event. These events are then excluded from the current concurrent merging set. For change events confirmed not to be duplicates by the idempotency check, domain change evidence records are generated and fused sequentially according to evidence time order and preset priority rules. Since clocks on distributed deployments may be inconsistent or have the same timestamp, each domain change evidence record also carries a monotonically increasing write sequence number assigned by the dynamic semantic state field. When evidence times are the same or incomparable, the write sequence number determines the order, ensuring the merging order is uniquely determined in all cases. The two baseline identifiers carried in the domain change evidence record are used for two types of failure determination: When the baseline semantic version identifiers are inconsistent, it indicates that the semantic state on which the change is based has been changed by prior fusion, so the promotion and conflict determination are re-executed on the current semantic version; when the baseline model version identifiers are inconsistent, it indicates that the domain model version on which the change is based has been changed by prior incremental reprojection or other changes. In this case, the original model difference cannot be directly applied to the current model version, so the difference is first recalculated by performing a rebasing or three-way comparison relative to the current formal domain model version, and then the atomic model change is decomposed again and semantic change candidates are generated accordingly. The two types of determinations are independent of each other, and any inconsistency triggers the corresponding reprocessing. For the same unified semantic projection range affected by multiple non-repeating change events, the scope of influence determination and incremental reprojection are only uniformly executed once after the evidence fusion is completed; this incremental reprojection adopts the aforementioned cross-target atomic transaction mode, which includes the model patches of each affected domain model, the update of the corresponding projection relationship records, and the comparison of semantic conservation quantities into the same logical transaction participation set.When all participating targets are successfully prepared and all semantic conservation comparisons pass, a unified commit is made. If any participating target fails to prepare, the model patch application fails, or the semantic conservation comparison fails, all participating targets are rolled back to the model version and projection relationship record version before the patch execution, maintaining an isolated state where their synchronization eligibility has been revoked. The boundary of this transaction is limited to the domain model side; the evidence fusion results and unified semantic object version already written are not rolled back due to the failure of this transaction. The reason for this definition is that the advancement of the semantic state is determined by evidence fusion and convergence criteria; if the semantic state is rolled back due to the failure of the domain model side patch, the accepted changes will be discarded, which contradicts the append-only immutable evidence discipline. Placing the failed model in an isolated state and waiting for the scope of influence determination and incremental reprojection to be re-executed based on the current unified semantic state version avoids intermediate states and repeated propagation of some applications, and also avoids the old version model whose semantic benchmark has changed from continuing to participate in synchronization.

[0091] (viii) Replay-based refusion and import of existing models.

[0092] Replay-based re-fusion. When the association between semantic evidence fragments and unified semantic objects is corrected—for example, when subsequent evidence shows that a fragment in a semantic evidence record actually describes another object—the system will deassociate the corresponding semantic evidence fragment from the original associated unified semantic object and re-establish the association. Both deassociation and re-establishment are achieved by appending association event records. The original association records, original semantic evidence records, and original fusion results are not deleted or rewritten to maintain the integrity of the audit chain. Each time evidence fusion forms a determined semantic state, the system records the evidence set identifier supporting the determined semantic state, the semantic fusion configuration version, and the fusion result version in the evolution field. The semantic fusion configuration version includes at least the anchoring algorithm and threshold used to associate and match semantic evidence fragments with unified semantic objects, the evidence fusion algorithm, the version of pre-set prior knowledge, the rules for calculating the reliability of evidence sources and the snapshot of the source reliability weights used in this instance, the version of constraint rules, the convergence threshold, the confidence difference threshold, the number of stable updates, the conflict arbitration rules, the evidence related groups and discount weighting rules, and the rules for generating, deduplicating, invalidating, and weighting restricted system candidates. After the association relationship is corrected, the system performs a replay-based refusion on all semantic content of the affected unified semantic object determined by the evidence, based on the immutable semantic evidence records, semantic evidence fragments, additional association events, and corresponding semantic fusion configuration versions stored in the evidence field. The reconstructed content includes semantic state, attributes, relationships, behavioral capabilities, constraints, constraint satisfaction states, and the confidence level of candidate semantic hypotheses, rather than just reconstructing conceptual hypotheses. Semantic evidence fragments, rather than entire semantic evidence records, are used as the anchor unit for modification, allowing an image, requirement document, or existing model file describing multiple objects to be partially anchored without affecting the remaining fragments. The replay-based refusion includes a historical configuration replay mode and a current configuration re-evaluation mode. The historical configuration replay mode reads the semantic fusion configuration version corresponding to the semantic state to be replayed, the source reliability weight snapshot, and the then-currently valid constraint rules. Fusion is re-executed under the condition that the fusion rules remain unchanged, used to distinguish the result changes caused by the association relationship correction itself. For restricted system candidates, only candidate auxiliary contributions whose validity can be confirmed based on the candidate identifier, generator version, baseline semantic version, and historically valid state are restored; those whose historically valid state cannot be confirmed are not included in the historical configuration replay. The current configuration re-evaluation mode uses the currently valid semantic fusion configuration version to re-perform fusion on relevant historical semantic evidence. It is applicable to re-evaluation after changes in fusion algorithm, source reliability model, prior knowledge, constraint rules, or convergence rules.The current re-evaluation does not overwrite the previously determined semantic state, the original fusion result, or its semantic fusion configuration version. Instead, it records the reason for the re-evaluation, the old and new semantic fusion configuration versions, the differences in results, and the scope of impact as a new re-evaluation event, and generates a new unified semantic object version. Restricted system candidates that are no longer valid or do not match the baseline semantic version are not directly used; instead, they are regenerated or re-verified by the current generator version and then used to calculate the restricted confidence of candidate semantic hypotheses. After replay, the system determines the replay result according to the same convergence conditions, determination thresholds, and hard constraint gating conditions as ordinary evidence fusion. If the replay result differs from the current valid semantic state, and satisfies the convergence condition, reaches the corresponding defined threshold, and has no unresolved hard constraint conflicts, a new accepted semantic change is formed, and a new unified semantic object version is generated. If the replay result shows that the originally determined semantic content no longer satisfies the convergence condition, no longer reaches the corresponding defined threshold, or no longer satisfies the hard constraint gating condition, the system generates a determined semantic withdrawal event in an append-only immutable manner. In the new unified semantic object version, the corresponding semantic information item is changed from the determined semantic state to a candidate semantic hypothesis state, an undetermined state, or a conflicting state, while retaining the original determined semantic content, the original fusion result, and its historical versions. If the replay result does not change the current determined semantic state, the current relation state, the current constraint state, or the target projection completeness, the unified semantic state change event is not generated again, or incremental reprojection is performed. When the replay result results in a new accepted semantic change, a confirmed semantic withdrawal event, or causes a change in at least one of the relation state, constraint state, or target projection completeness, the system determines the formal domain model that depends on the affected unified semantic object based on the projection relationship record, temporarily revokes its synchronization eligibility, and recalculates the completeness of the corresponding target projection. If the recalculated target projection still meets the preset model generation conditions, the system determines the scope of influence, performs incremental reprojection, and restores its synchronization eligibility after the domain model obtained by incremental reprojection passes the semantic conservation quantity verification. If the corresponding target projection no longer meets the preset model generation conditions due to confirmed semantic withdrawal, the corresponding formal domain model remains in an isolated state where its synchronization eligibility has been revoked and is marked as semantically incomplete. The system enters the active completion, draft generation, or regeneration process. Its synchronization eligibility is not restored until the target projection again meets the preset model generation conditions and the regenerated or incrementally reprojected domain model passes the semantic conservation quantity verification. Other formal domain models that do not depend on the affected unified semantic object do not lose their synchronizability due to this replay, thereby keeping the scope of reversal consistent with the actual scope of impact and allowing erroneous associations to be corrected while preserving the complete historical record.

[0093] Existing model import. Existing domain model files can be used as semantic input data. The lifting operator, registered with the corresponding target domain, processes the data in two phases using an initial import mode. In the first phase, the lifting operator parses the model elements, relations, constraints, and attributes containing object semantic content from the existing domain model file, and generates semantic evidence fragments for those that can constitute independent semantic processing units. These model elements, relations, constraints, and attributes that can constitute independent semantic processing units are collectively referred to as imported semantic units. Each semantic evidence fragment carries independent location information, candidate semantic hypotheses, and the source handle of the imported semantic unit. Connections, dependencies, inclusions, or other type-related relationships between model elements generate semantic evidence fragments indicating relational semantic elements; constraint expressions and their binding relationships generate semantic evidence fragments indicating constraint semantic elements; attributes that can be independently referred to or independently valued generate semantic evidence fragments indicating those attributes; inline relations, inline constraints, and inline attributes that cannot constitute independent semantic processing units are carried along with the semantic evidence fragments of their respective model elements. Information such as canvas coordinates, primitive styles, view collapse states, annotations, and editor auxiliary elements that do not carry object semantic content are not generated into semantic evidence fragments to avoid propagating information used only for display or editor operations as semantic elements across domains. Semantic evidence fragments are generated for relations, constraints, and attributes that can be independently referred to or independently valued, enabling them to be anchored to relational unified semantic objects, constraint unified semantic objects, or attribute semantic units, and to independently perform anchoring, replay, projection relationship establishment, and reverse promotion. The first stage is executed according to either a full import mode or an incremental import mode. The full import mode parses all imported semantic units with object semantic content in the existing domain model file; in one implementation, by comparing the existing domain model file with an empty baseline, a full set of newly added differences covering the import scope is obtained, and the full set of newly added differences, along with the set of semantic evidence fragments generated by directly parsing all imported semantic units, are used to parse only the model differences of the existing domain model file relative to a preset non-empty baseline model. The incremental import mode is only used when the following conditions are met simultaneously: the preset baseline model has been incorporated into the unified management of the dynamic semantic state field; each imported semantic unit that is shared by the preset baseline model and the existing domain model file and can constitute an independent semantic processing unit has already had the object identifier and corresponding projection relationship record of the unified semantic object associated with it; the shared imported semantic units include model elements, relations, constraints and attribute items that can independently take values ​​or be independently referred to, which have object semantic content; for inline relations, inline constraints and inline attributes that do not constitute independent semantic processing units but are carried along with the model elements to which they belong, their semantic content summary and version identifier have been incorporated into the unified management of the baseline model along with the imported semantic units to which they belong.When any shared imported semantic unit lacks the object identifier, projection relationship record, semantic content summary, or version identifier of its associated unified semantic object, a full import is performed on the existing domain model file, or at least a supplementary full import is performed on the imported semantic unit category lacking a unified management relationship. This is because, in incremental import mode, shared parts do not generate new semantic evidence fragments; if shared relationships, constraints, attribute items, or model elements have not yet entered the dynamic semantic state field, only the differing parts can obtain unified semantic objects and projection relationship records, and the shared parts cannot be uniformly managed, making subsequent reverse promotion of these parts impossible to locate. In the second stage, the aforementioned semantic evidence fragments are associated with existing unified semantic objects according to the evidence object anchoring rules, or new entity-type, relation-type, or constraint-type unified semantic objects are created, and they participate in fusion according to the evidence fusion rules. After anchoring, a temporary projection relationship record is established between each imported semantic unit and its associated unified semantic object based on the imported semantic unit source handle carried by each semantic evidence fragment. The source handle of a model element can be taken from its path or identifier in the existing domain model file; the source handle of relations and constraints can be taken from the path or identifier of their independent elements in the file; the source handle of an attribute item can be taken from the path or identifier of its parent model element and a combination of the attribute field identifier. For inline relations, inline constraints, and inline attributes carried along with their parent model elements, their dependency references, semantic content summaries, and version identifiers are recorded in the temporary projection relation record of the parent imported semantic unit. The order of the two stages is not reversed. The projection relation record uses the unified semantic object identifier or attribute item identifier as a necessary field. Before the semantic evidence fragment has been anchored and the corresponding unified semantic object has been determined or created, there is no canonical semantic identifier that can be referenced by the temporary projection relation record. Therefore, the temporary projection relation record is established after anchoring is completed. For an existing domain model to achieve a synchronized formal domain model state, the following three conditions must be met simultaneously. First, the semantic information items involved satisfy the convergence condition and reach the corresponding determined threshold. Second, the semantic information item has independent source semantic evidence other than the existing domain model file, is confirmed by the user, is verified by independent deterministic constraints, or the existing domain model file is explicitly marked as a trusted initialization source. The marking of a trusted initialization source has both trust and integrity basis. Trust basis includes at least one of the following: confirmation by a subject with preset permissions; verification through a trusted digital signature or source certificate; confirmation through a preset approval process; or belonging to a pre-registered authoritative model library. Integrity basis includes at least file content digest verification, version consistency verification, and complete source record. Files that only complete content digest verification and version consistency verification but lack trust basis do not constitute a trusted initialization source. The marking subject, marking basis, file content digest, confirmation time, and scope of application are recorded in the evolution field. Third, the semantic conservation quantity verification passes.Semantic conservation quantity verification is performed in one of two ways. The first way involves generating the expected semantic conservation quantity or verification expected model for the target domain based on the updated unified semantic projection range. The existing domain model itself is then used as the object of verification for semantic conservation quantity comparison, and the comparison does not overwrite the existing domain model with the generated result. When the consistency deviation is within the corresponding threshold, the existing domain model achieves a synchronizeable formal domain model state. Temporary projection relationship records corresponding to each imported semantic unit in the existing domain model file are converted into formal projection relationship records according to the convergence state, independent verification results, and semantic conservation quantity verification results of their corresponding semantic information items. For inline relationships, inline constraints, and inline attributes carried along with the model elements, their dependency references, semantic content summaries, and version identifiers are retained in the formal projection relationship records of the imported semantic units. The second way involves re-performing the formal projection based on the updated unified semantic projection range to generate a new domain model for the target domain. For the model elements, relationships, constraints, and independent attribute items in the generated model, projection relationship records to be verified are first established, and then semantic conservation quantity comparison is performed on the generated domain model. When the consistency deviation is within the corresponding threshold, the domain model generated by the reprojection is used as the formal domain model for the target domain, and the corresponding projection relationship records to be verified are converted into formal projection relationship records. Temporary projection relationship records pointing to the imported semantic units of the original existing domain model are retained as import source relations and are not converted into formal projection relationship records. The two methods are not mixed. When using the second method, the semantic conservation quantity verification verifies the domain model generated by the reprojection. The temporary projection relationship records pointing to the original existing domain model cannot be converted into formal projection relationship records based on the verification result. Otherwise, the verified object and the object that has obtained formal mapping qualification belong to two different models, and the model element identification system may also be different. Before all three conditions mentioned above are met, the existing domain model is only marked as the import baseline model. Its content can be browsed by users and they can continue to supplement semantic evidence, but it does not obtain a synchronized formal domain model state. When the semantic conservation quantity verification fails using the first method, an additional verification failure event record is added, and the existing domain model does not acquire a synchronized formal domain model state; the temporary projection relationship record corresponding to each imported semantic unit in the existing domain model file is marked as failing formalization and remains in the temporary projection relationship record state; after supplementing semantic evidence, completing independent verification, or correcting the unified semantic projection range, the verification can be re-executed.When the semantic conservation quantity verification fails using the second method, the domain model generated by the reprojection does not acquire a synchronizeable formal domain model state. The projection relationship records established for the model elements, relationships, constraints, and independent attribute items in the generated model are marked as verification failed and are not converted into formal projection relationship records. Temporary projection relationship records pointing to the imported semantic units of the original existing domain model continue to be retained as import source relationships and are not marked as invalid due to the failure of this reprojection verification. None of the above historical records are physically deleted. The independent verification or trusted initialization condition is set because when all semantic evidence for a certain semantic information item comes from the same existing domain model file, the baseline semantic conservation quantity formed by that semantic information item is from the same source as the file being verified. In this case, the conservation quantity comparison mainly verifies whether the round-trip link composed of the projection operator and the conservation quantity extraction logic is faithful, and cannot independently prove that the content of the existing model is faithful to the real object to be modeled. This rule, together with the rule that a determined semantic state cannot be obtained when all candidate support sources are restricted system candidates, constitutes a restriction on self-verification from a single source. The initial import mode and the domain model change promotion mode can share the same promotion operator interface. The initial import mode takes the entire existing domain model file or the model difference relative to the managed baseline model as input, and does not rely on existing formal projection relationship records during the initial import. The change promotion mode takes the model difference of the formal domain model as input and relies on existing formal projection relationship records and projection instance handles for positioning. The semantic evidence records generated during import and semantic evidence records formed with other information sources participate in evidence fusion according to the same evidence object anchoring, evidence relevance groups, source reliability, constraint gating, and convergence rules, thereby supporting the unified semantic integration of legacy model data from multiple tools.

[0094] (ix) Adaptation and security filtering of external artificial intelligence models.

[0095] External artificial intelligence models (visual recognition models, large language models, multimodal models, etc.) are accessed as one of the information sources through the input adapter of the access and interaction layer: the candidate interpretations output by the model are mapped to candidate semantic hypotheses; the confidence information output by the model is converted into the initial evidence support after being converted by the normalization or calibration rules corresponding to the model. s k ，The normalization or calibration rules are configured according to the model identifier and version respectively; when the model does not provide confidence information, the default support value is taken according to the preset default support value of the information source type, and the support value is recorded as the default value in the source information; the above content, together with the source information, timestamp and evidence quality score, constitutes the semantic evidence record entry. The output of the external artificial intelligence model does not directly modify the semantic state, relationship or constraint of the unified semantic object. The reason for setting the conversion and default value rules is that the score units and calibration methods of different models are different. If they are directly substituted into formula (5) as support, the reliability weight index will be affected. α i Loss of comparability. The reliability weights of the evidence sources corresponding to the external artificial intelligence model are dynamically adjusted based on its historical semantic parsing accuracy. During statistics, the semantic conclusions determined by the convergence criterion and not refuted by subsequent evidence are used as the correctness benchmark. When the number of samples is less than the preset minimum sample size, the default reliability of the information source type is used, or the default reliability is used as a priori and the value is taken in a Bayesian smoothing manner, rather than directly drawing a distortion evaluation based on individual samples. When the identifier, model version, prompt configuration, or output parser of the external artificial intelligence model changes, its historical reliability is reset to the default value in a preset manner, continues to accumulate after being converted by a preset attenuation coefficient, or is saved separately by version without being carried over across versions, so as to avoid evaluating the new version with the performance of the old version. The dynamically adjusted weight values ​​are recorded as a source reliability weight snapshot in the semantic fusion configuration version corresponding to the fusion each time they participate in fusion, so that the fusion result can be reproduced based on the original weight snapshot.

[0096] The security filter is not a standalone filter component, but rather consists of three existing mechanisms: firstly, source reliability weighting, where the reliability weighting index of the artificial intelligence source in equation (5) is... α iThe system is configurable and dynamically adjusted based on historical accuracy. Secondly, it integrates convergence criteria. Before satisfying equation (6), the unilateral assertion of artificial intelligence is only a candidate hypothesis and does not enter the determined semantic state. Thirdly, it employs dual gating. Even if it is accepted, it still needs to pass the completeness gating of equation (9) and the conservation gating of equation (11) to affect the formal output. In addition to the above three mechanisms, the candidate semantic hypotheses generated by the external artificial intelligence model output are subject to the same hard constraint diversion rules as the semantic change candidates in Part (VII): those that violate the non-exemptible hard constraints after deterministic evaluation are placed in the rejection state and do not participate in fusion or enter the determined semantic state; those that cannot be determined whether they violate the non-exemptible hard constraints are placed in the conflict state and do not participate in fusion until the conflict is resolved; those that violate the exemptable hard constraints after deterministic evaluation but have not yet completed the effective exemption are also placed in the conflict state. After completing the effective exemption, they are re-evaluated. If the exemption is rejected or exceeds the preset exemption processing conditions, they are converted to the rejection state; those that violate the soft constraints suppress their confidence according to the corresponding penalty coefficient. The reason for using the same triage rule in both places is that the acceptability judgment of hard constraints should not differ based on the candidate source; otherwise, the same value proposed by artificial intelligence and proposed by model editing will be treated differently, and hard constraint gating will lose its determinism. If external models are allowed to output directly written semantic states, model illusions or errors will pollute all downstream domain models without traceability. Under this scheme, erroneous outputs only manifest as a low-support evidence, which can be overwhelmed by subsequent evidence and can be corrected losslessly through replay-based refusion. External artificial intelligence models can also serve as active completion collection actions. act When invoked, its return also enters the scene in the form of evidence.

[0097] (x) Restricted participation of candidates in restricted systems.

[0098] The system can generate historical prediction candidates, correction suggestion candidates, and repair proposal candidates, collectively referred to as restricted system candidates. Restricted system candidates are not included in the number of evidence items in equation (5) as semantic evidence records, nor are they included in the independent evidence related group or semantic evidence source category. Instead, they serve as independent candidate auxiliary factors, influencing the confidence level of the corresponding candidate semantic hypothesis with restricted weights. Let the effective weight of the restricted system candidate be... α lim The reference weight is α ref The preset limited attenuation coefficient is r Then the effective weights of the candidates in the restricted system satisfy equation (14): (14) In equation (14), α lim Represents the effective weights of candidates in a constrained system; α ref Indicates the reference weight; r This indicates a pre-defined, constrained attenuation coefficient. When there is actual observational evidence from an unconstrained source for the same semantic information item... α ref Take the minimum value among the effective weights of unrestricted source semantic evidence participating in the calculation of this semantic information item; when there is no actual observational evidence from unrestricted sources. α ref Take the preset restricted benchmark weights α 0, in this case, the restricted system candidate is determined by... α 0 and r The restricted weights determined by Equation (14) are used in the calculation. These restricted weights are only used to maintain the relative order of candidate semantic hypotheses and do not allow the corresponding semantic information items to obtain a determined semantic state. For restricted system candidates with the same baseline semantic version, the same generator version, the same target semantic information item, and semantic equivalence, the system performs replacement, merging, or discount weighting without repeated accumulation. The generation or update of restricted system candidates is not included in the number of updates of independent evidence-related groups in the stable window of Equation (6). When all candidate support sources for the same semantic information item are restricted system candidates, the semantic information item cannot obtain a determined semantic state. It can only obtain a determined semantic state after obtaining semantic evidence from unrestricted sources, or after passing independent constraint solving, simulation, or deterministic verification and generating a semantic evidence record accordingly. Restricted system candidates are only used to assist candidate reasoning and confidence calculation. They cannot replace semantic observation, nor can they trigger formal domain model generation, synchronization, or incremental reprojection on their own.

[0099] (xi) Determination of the target projection set and mission version.

[0100] Target projection set G The target projection set is determined by at least one of the following methods: receiving user specification; reading preset configuration; and adding target projections that meet the preset recommendation conditions to the target projection set based on the degree of matching between semantic information items that have met the convergence conditions in the dynamic semantic state field and the semantic information requirements declared by each registered target projection (preferably using the sub-completeness of equation (7) as the matching metric). To avoid looping, the determination of the target projection set and the formation of the unified semantic projection range are performed in two stages: the first stage determines the candidate semantic range in the dynamic semantic state field by the root unified semantic object, the explicit specification of the current modeling task, or the user selection, and calculates the degree of matching within the candidate semantic range, generates a recommended target projection list, and determines the target projection set for this round and generates a task version identifier; the second stage forms the unified semantic projection range corresponding to the determined target projection set. VThis serves as the processing unit for determining completeness, generating projections, comparing semantic conservation quantities, and determining the scope of influence. The system can automatically analyze "which domain models the current semantics are sufficient to generate" and proactively recommend models; users can also add or delete them. To prevent the target projection set from drifting during the same round of task execution, the system configures the target projection set and its settings for this round (the semantic information items declared by each target projection). s d Weight w v With determining the threshold i v det The task version identifier is generated and remains unchanged during the completeness determination, projection generation and consistency verification of this round. During execution, new recommendations triggered by the matching degree are only written into the recommended target projection list and merged into the next task version after user confirmation or preset strategy judgment. If the target projection set is allowed to expand in the middle of this round, there will be repeated triggering of "completeness meets the standard - new target is added - completeness falls back - active completion again". This design, along with the stable window condition of equation (6) and the echo suppression of system source update, belong to the bounded design of the feedback loop of this invention. The content frozen in the task version is not limited to semantic information items, weights, and determination thresholds, but also includes the versions of projection operators registered in each target domain, the versions of boosting operators, the versions of conservation quantity extraction configurations, the versions of normalization rules, and the deviation thresholds and generation and draft thresholds of each conservation quantity component. If any of the above versions changes during this round of execution, the change will not be applied to this round of tasks, but will be incorporated into the next task version along with the new recommendation. This ensures that the configuration on which completeness calculation is based, the operators on which projection generation is based, and the extraction configurations and normalization rules on which semantic conservation quantity comparison is based correspond to each other under the same task version, and the task results can be deterministically reproduced.

[0101] (xii) Beneficial effects.

[0102] Compared with existing technologies, this invention has at least the following beneficial effects: the above-mentioned links are not independent functional superpositions, but are interdependent at both the data and constraint levels. The appended immutable storage of the evidence layer provides a traceable input basis for anchoring correction and replay-based refusion; the determination result of the convergence criterion constitutes the input for the per-objective completeness calculation; the completeness gating determines whether and in what state projection is performed; the projection relationship records established during the projection process constitute the basis for calculating the scope of influence in the change reflow stage; and the semantic conservation gating determines whether the domain model can obtain synchronization eligibility and thus enter the change reflow stage. Because of the above-mentioned dependencies, the cross-domain semantic consistency, end-to-end traceability, and synchronization stability generated by the combined action of each link are technical effects that cannot be achieved by implementing any one of the links alone.

[0103] A three-tiered gating system ensures output quality: convergence criteria answer "is the semantics definite?", completeness answers "is the information sufficient?", and conservation checks answer "is the generated product faithful?". These three gating systems are sequentially prioritized. Simultaneously, the reverse synchronization link incorporates a propagation gating system that ensures convergence and the absence of unresolved hard constraint conflicts. Any changes undergo semantic stability checks before cross-domain propagation, preventing the spread of erroneous semantics. Any domain model passes semantic conservation checks before gaining synchronization eligibility. Draft domain models, after passing structural validity and non-exemptible hard constraint checks, are presented as drafts and do not gain synchronization eligibility. Neither type of model enters the synchronization link without verification. For insufficient information, there are flexible options for draft downgrading and proactive completion, preventing the silent outflow of incomplete or distorted models.

[0104] Concurrent changes are securely mergeable: Domain change evidence records carry a baseline semantic version identifier, a baseline model version identifier, and an idempotency identifier. When the baselines are inconsistent, promotion and conflict determination are re-executed on the current semantic version. Duplicate changes are discarded after being identified by the idempotency identifier. For multiple change events that jointly affect the same unified semantic projection range, only one unified determination of the impact range and incremental reprojection are performed after fusion. Concurrency control is thus moved from the file or model element level to the semantic level, avoiding both overwrite loss and duplicate propagation, and enabling cross-domain semantic mutual exclusion (each being legal but violating hard constraints after merging) to be detected by the constraint field before propagation.

[0105] Synchronization complexity is linear: the number of target neighborhoods is n D Registration is required. n D Group operator plugins, cross-domain synchronization propagation via star topology, have a link complexity of O(n). n D ), compared to the O( ) of the point-to-point mapping architecture n D 2 ) Optimizes complexity hierarchy; adding, updating or removing any domain plugin does not change the core data structure of the dynamic semantic state field and the unified semantic object; performs compatibility checks, failure marking or reprojection on projection relationship records and domain models affected by operator version changes, while other unaffected target projections remain unchanged.

[0106] Trustworthy and traceable learning with secure learning: Any model element can be traced back to semantic evidence fragments and their sources through projection relationship records and formal bidirectional indexes; external artificial intelligence models only enter the evidence layer and are filtered by three mechanisms, and their capabilities are securely absorbed; the system achieves continuous learning and self-correction through evidence accumulation, anchoring correction, and replay-based re-fusion, and what is maintained is the activity mechanism of the semantic field rather than a static knowledge base.

[0107] All final states are observable: the formal domain model, the draft domain model, incompleteness prompts, and consistency check failure prompts all reach the user through the access and interaction layer. The system has no silent dead zones, and failures can be located, handled, and restarted with supplementary input. Attached Figure Description

[0108] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments are briefly introduced below. Obviously, the following drawings are only some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0109] Figure 1 The schematic diagram of the cross-domain modeling method and system architecture for unified semantic objects provided in the embodiments of the present invention shows the semantic object association and fusion module A1, evidence generation module A2, change evidence and improvement module A3, dynamic semantic state field A4 (including entity field, relation field, constraint field, evidence field, evolution field, hypothesis field), completeness determination module A5, influence analysis and incremental reprojection module A6, projection and consistency verification module A7, as well as the access and interaction layer B1 and domain model B2 as the system boundary. In this process: B1 provides semantic input data to A2; A2 records semantic evidence in the evidence field of A4; A1, A5, A7, A3, and A6 interact bidirectionally with A4; A5 sends a supplementary request to B1 when completeness is insufficient; A5 triggers A7 when completeness is satisfied; A7 is the sole projection execution point for the domain model, generating a formal or draft domain model for B2 via projection operators; B2's change events are sent to A3; A3, after being evidenced by the boosting operator, writes the data into A4; A6, after determining the scope of influence, sends an incremental reprojection request to A7. The projection operators registered in each target domain are marked on the line connecting A7 and B2 in the diagram. P d Mark the lifting operators registered in each target domain on the line connecting B2 to A3. L d In A7, annotate the conservation quantity extraction logic Ext corresponding to each target domain. dAll three are registered as domain plugins and can be added, updated, or removed independently. B1 and B2 are bidirectionally connected: B1 to B2 represent the domain model editing operations performed by the user at the access and interaction layer, acting on the corresponding domain model; B2 to B1 represent the formal state domain model, draft state domain model, and their state markers presented to the user via the access and interaction layer. User input at the access and interaction layer is categorized into distinct pathways: semantic input provided in the form of natural language, images, files, parameter values, or decisions regarding candidate states awaiting user confirmation passes through B1 to A2, generating semantic evidence records; model editing operations provided in the form of adding, deleting, modifying attributes, or adjusting associations of domain model elements pass through B1 to B2, and after B2 issues a model change event, A3 applies a boosting operator to generate domain change evidence records; both types of records are written to A4 and participate in the same evidence fusion process. The control configuration inputs provided in the form of thresholds, target projection sets, and their configurations directly update the running configurations used by each module and the task version configurations of this round of modeling tasks without going through A2. They are not used as semantic evidence records in semantic fusion. The product written back to B2 by incremental reprojection through A7 carries the system source update marker. A3 does not generate domain change evidence records for model change events carrying this marker, thereby suppressing echoes.

[0110] Figure 2 This is a flowchart illustrating the cross-domain modeling method for unified semantic objects provided in this embodiment of the invention. It shows steps S1 to S5, the completeness determination branch (when not satisfied, a supplementary request is executed through the access and interaction layer or an incomplete state is presented), the consistency determination branch (when repairable, a repair proposal is executed and re-compared; when the repair termination condition is met, a repair proposal or verification failure details are presented), the path of the formal domain model and incremental reprojection product presented through the access and interaction layer, and the dual role of the access and interaction layer in providing input and presenting results. Figure 2 The main timing of the method is shown. The complete internal branches of each step are subject to the text of the specification. The detailed steps of the draft state domain model generation branch, target-by-target projection gating, establishment of projection relationship records, state diversion of semantic change candidates, temporary cancellation and restoration of the synchronous qualification of the source domain model, semantic conservation quantity closed-loop verification of the source domain model, system source update marking and echo suppression, active completion termination conditions and repair loop omitted in the figure are respectively found in Parts (V), (VI) and (VII) of the invention content and specific implementation of the specification. The figure does not constitute a limitation on the above content. Detailed Implementation

[0111] Preferred embodiments of the present invention are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention; the module names, weights, and threshold values ​​used herein are merely examples for ease of explanation. Any changes or substitutions made to the embodiments without departing from the principles of the present invention fall within the scope of protection of the present invention.

[0112] Example 1: Cross-domain modeling of intelligent driving domain controllers.

[0113] This embodiment is based on Figure 1 The architecture shown, by Figure 2 The process shown takes the intelligent driving domain controller as the object and the target projection set. G It encompasses four target areas: 3D structural modeling (assembly models that can be viewed individually by component), hardware architecture modeling (board and bus topology diagrams), system logic modeling (SysML internal block diagrams and state machines), and software architecture modeling (UML component diagrams). According to Part (XI) of the invention, the system automatically recommends two target projections: hardware architecture and SysML, based on the converged semantic information items. Engineers then add selections for 3D structure and UML through the access and interaction layer, forming the target projection set for this round and generating a task version identifier.

[0114] Engineers provided five semantic input data streams: a textual requirement: "Design an L2+ intelligent driving domain controller, connecting 8 cameras and 2 CAN-FDs, using a certain type of SoC as the main computing chip, with a total power consumption not exceeding 35 watts, and supporting fault diagnosis and reporting"; a photo of a reference board layout; a CAD file of an existing structural component; an existing CAN driver header file; and the parsing results of the requirement text from an external large language model and the parsing results of the photo from an external visual model. Thresholds and weights were set via control configuration inputs, without going through an evidence layer.

[0115] Step S1: Evidence generation module A2 generates semantic evidence records according to formula (3). e 1 (Source: Requirements text parsed by a large language model) α 1=0.8): Contains a segment describing the whole machine, candidate hypothesis {domain controller: 0.90}, attributes "number of cameras = 8, number of CAN-FD channels = 2, power consumption limit = 35 W", behavior capability "fault diagnosis and reporting"; e 2 (Source: Photo analyzed by a visual model) α2=0.7): Contains five semantic evidence fragments, located in the SoC chip area, the first row of connector area, the second row of connector area, the heat sink outline area, and the independent small box area on the right side of the screen, respectively; the candidate hypotheses for the two connector fragments are {camera connector: 0.75, Ethernet connector: 0.40}, and the candidate hypotheses for the independent small box fragment are {independent ECU: 0.50, power filter module: 0.45}; e 3 (Source: CAD file parsing) α 3=0.95): Shell geometric envelope 260×180×45 mm, mounting hole positions and positioning reference; e 4 (Source: Code header file parsing) α 4=0.92): Interface signature "CanFdDriver::send (frame)", baud rate parameter; e 5 (Source: Visual model's OCR region analysis of the photo) α 5=0.6): The nameplate area appears to contain the characters "40 W". Five pieces of evidence were added to the evidence field in an additive, immutable manner. The parsing of the CAD file and header file was performed in initial import mode using the lift operator registered in the corresponding target domain, which falls under the category of importing existing models as described in Part (VIII) of the invention. The scores output by the external model were converted into initial evidence support using their respective calibration rules. s k The CAD files and code header files are taken from the enterprise's controlled document library. After being confirmed by a subject with preset permissions and passing the file content summary and version consistency verification, they are explicitly marked as trusted initialization sources by the control configuration input. The marking subject, marking basis, content summary and confirmation time are recorded in the evolution field. Therefore, the temporary projection relationship record established by its import is converted into a formal projection relationship record after the relevant semantic information items satisfy the convergence criterion of equation (6) and the domain model obtained by reprojection passes the semantic conservation quantity verification. Existing model files that are not marked as trusted initialization sources or whose marking does not have the above verifiable basis must wait for the semantic information items involved to obtain independent source semantic evidence, user confirmation or independent deterministic constraint verification outside the file before their temporary projection relationship record can be formalized.

[0116] Step S2 (Anchoring): The semantic object association and fusion module A1 performs evidence object anchoring according to formula (4) on a unit basis, using semantic evidence fragments. In this embodiment, we take... m 1 = 0.5 m 2 = 0.3 m 3 = 0.2, correlation threshold i anc =0.75, anchoring margin threshold i mar =0.10.e 1. Arrive first, the entity field is empty, create a unified semantic object. U 1 (Domain Controller); e In Figure 2, the SoC chip segment, the first row of connector segments, the second row of connector segments, and the heatsink outline segment are relative to each other. U The anchoring scores for 1 were 0.88, 0.84, 0.81, and 0.86, respectively, all of which were no lower than [the required scores]. i anc And the difference between it and the second-highest score is no less than i mar Therefore, they are all determined to be associated with U 1; e The independent small box fragments in 2 and U The anchoring score of 1 is 0.42, which is lower than... i anc Create a new object U 2. Its candidate hypothesis {Independent ECU: 0.50, Power Filter Module: 0.45} has not converged and continues to exist in the hypothesis field as a multiple hypothesis state; e 3. e The fragment in number 4 is similarly related to... U 1. The scope of this round of unified semantic projection is denoted as... V 1. Includes U 1. U 2. and its entityized relational objects and associated constraint objects; among which U The object identity of 2 has not yet satisfied the convergence criterion of equation (6), and its inclusion V 1. For carrying purposes only U 1. The relationships and constraints between it and other objects. U 2. It is not a required semantic element for any target projection in this round, does not participate in the generation of this round of projection, and its object identity is not included in the object identity set of the baseline semantic conservation quantity.

[0117] Step S2 (fusion and conflict arbitration): A1 Perform logarithmic domain incremental fusion according to equation (5). U The confidence level of the "domain controller" hypothesis after three rounds of updates is 1. The power consumption reached 0.95; a conflict occurred in the power consumption attribute. e 1. Advocate 35 W ( α 1=0.8) e 5. Advocate 40 W ( α (5=0.6), the total variation distance of the support distribution exceeds the conflict threshold. i conf=0.6, weighted by source reliability, the arbitration tendency is 35 W, and a manual review item is generated and presented through the access and interaction layer B1; the engineer confirms 35 W (the nameplate shows the maximum instantaneous power consumption), and the review result serves as evidence of high reliability. α =0.98) additional field re-fusion. This embodiment uses convergence criterion parameters. t 1 = 0.85 t 2 = 0.20 n stab =3, U The core semantic information items of 1 converge in equation (6) one by one. In the relation field U 1 and U The power supply relationship between the two is materialized into an independent semantic object "Power Supply Relationship #1", carrying current constraints and supporting evidence references. After fusion, the dynamic semantic state field A4 updates the initial index to a formal bidirectional index.

[0118] Step S3 (Completeness and Active Completion): Completeness determination module A5 calculates according to equations (7) and (8), and takes the projection of each target. c gen,d =0.95、 c draft,d =0.50. The set of semantic information items declared by the four target projections. s d These are: 3D structure {geometric envelope, positioning reference, mating relationships}; hardware architecture {processing units, bus topology, interface list, power budget}; SysML logic {decomposed structure, port interfaces, state transitions}; UML software {component responsibilities, interface signatures, dependencies, timing parameters}. The sub-completeness Cmp is calculated. d ( V 1) With values ​​of 1.0, 1.0, 1.0, and 0.72 respectively, the "timing parameters" (frame processing period, diagnostic reporting period) of the UML projection did not converge, therefore Gen... UML =0 and the Gen of the projections of the other three targets. d =1. U Although the object identity of 2 has not converged, it does not belong to the set of semantic information items declared by the four target projections mentioned above. s d None of them are mandatory semantic information items. s d req It is also not a hard constraint related semantic information item s d hc Therefore, it is not included in the weighted statistics of equation (7) and does not affect the projection of each target Gen. dThe determination; accordingly, this round of active completion is only performed on the semantic information items (temporal parameters) to be completed declared by the UML projection that do not meet the generation conditions. U "Object identity" does not belong to any semantic information item to be completed in any target projection, so it will not enter the active completion in this round. Active completion is completion relative to the completeness of the target projection, rather than targetless completion of all semantic content. U The identity of 2 is left to the semantic evidence that arrives later (see the replay-based refusion example at the end of this specification). After the fusion satisfies the convergence criterion of equation (6), it is then incorporated into the corresponding domain model according to the influence range determination and incremental reprojection as described in Part (VI). Target-by-target projection gating is performed according to equation (9): The projection and consistency verification module A7 generates a draft state domain model for the UML projection and simultaneously establishes a draft projection relationship record with draft state identifiers, respectively recording the draft components and interface elements of CameraPipeline, CanFdDriver, and DiagnosticManager. U The corresponding behavior capabilities and attribute items of 1, the handle of each draft projection instance, the version of the projection operator, the version of the source semantics, the version of the draft model (denoted as draft-1) and the version identifier of this round of tasks; this record does not constitute a formal projection relationship record, but is only used for the location of draft elements and the improvement of draft changes. After the draft model passes the structural validity check (component and interface references are complete and there are no dangling dependencies) and the non-exempt hard constraint check (no interface protocol compatibility conflict), it is presented by B1 and marked as a draft. The other three target projections can be formally projected first; at the same time, A5 compares the candidate acquisition action set Act according to formula (10): "Asking the user for timing parameters" has an expected information gain of 1.8 bits and low cost, "Calling an external large language model to retrieve typical parameters by SoC model" has a gain of 1.2 bits and low cost but low reliability. The former is selected as act * B1 asked the engineer a question. The engineer replied, "Frame processing cycle 33 milliseconds, diagnostic reporting cycle 1 second," and the answer was confirmed by A2 through evidence-based entrance fusion, Cmp. UML ( V 1) Upgrade to 1.0, Gen UML Convert to 1.

[0119] Step S3 (Formal Projection): A7 uses the projection operators registered in each of the four target domains. P dPerform formal projection: 3D structure projection generates an assembly model that can be viewed individually by component (casing, SoC board, connector array, heatsink); hardware architecture projection generates a board topology diagram (SoC, 8-channel camera serializer / deserializer, 2-channel CAN-FD transceiver, power tree); SysML projection generates an internal block diagram (domain controller block with 8 camera input ports, 2 CAN-FD ports, power consumption attribute 35W) and a state machine (power-on → self-test → operation → fault reporting); UML projection generates component diagrams (CameraPipeline, CanFdDriver, DiagnosticManager components and interfaces). A7 simultaneously creates... V The projection relationship records between each semantic element in section 1 and the above model elements are recorded. None of the four domain models mentioned above contain any related semantic elements. U 2. Corresponding model elements, U 2. Only after the object's identity converges is it incrementally reprojected into the corresponding domain model. In the 3D structural model, individually viewable model elements such as "shell," "SoC board," "connector bar," and "heat sink" each correspond to a unified semantic object. U The component attributes of 1: The projection relationship record allows multiple model elements to be mapped to different attribute items of the same unified semantic object. In addition to the unified semantic object identifier and the model element identifier, the mapping record also records the identifier of the mapped attribute item. The mapping type is determined based on the attribute item. In this example, "casing", "SoC board", and "heat sink" each correspond to... U A component attribute item of 1 is denoted as a one-to-one relationship, while a "connector row" attribute item corresponding to multiple connector model elements is denoted as a one-to-many relationship. Therefore, when a user subsequently modifies a heatsink or connector row individually, the lift operator, based on the projection relationship, positions the corresponding atomic model change accordingly. UFor the corresponding attribute item of 1, generate semantic change candidates at the attribute item granularity, without degenerating into a global change for the entire unified semantic object; when a component attribute item converges independently with the support of subsequent semantic evidence, and its semantics can be independently identified as a component with an independent object identity (such as "SoC board" being confirmed by evidence as an independently procurable and individually changeable component), and has at least one of the following: independent projection relationship record, independently recordable evolutionary history, and independent constraint set, the attribute item is promoted to an independent unified semantic object through replay-based refusion, and the object identity set of the projection relationship record and the baseline semantic conservation quantity is updated accordingly. The objectification process for this attribute item is executed in the following order: First, a new entity-type unified semantic object is created and an independent object identifier is assigned; then, the determined semantic content, candidate semantic hypothesis groups and their confidence levels, constraint references and their satisfaction states, current convergence state, source semantic version, semantic evidence fragment references, evolution history, and projection relationship records originally associated with this attribute item are migrated or initialized to the created object. Migration can be achieved by directly copying the current determined semantic content and confidence snapshot of the attribute item and recording its source semantic version, or by performing a re-fusion with migration markers on the migrated semantic evidence fragments. Except for this re-fusion with migration markers, the migrated fragments are not counted again. If only evidence references are migrated without migrating the above states, the created object will exhibit a state break of "evidence references but no determined semantics, no initial value for the confidence function, and no migration of the original attribute item constraints," thus the semantic continuity before and after objectification is not established; subsequently... U 1. Establish a relationship object of the corresponding type (in this example, a composition relationship object) between the created object and the object. Finally, after objectification, when the object has been included in the corresponding unified semantic projection scope, and the corresponding target projection has declared the object's identity identifier as its conserved component and requires the generation of model elements corresponding to the object, the object's identity identifier is included in the object identity set of the baseline semantic conserved quantity corresponding to the target projection. Target projections that have not declared the generation of this model element (such as UML software architecture projections in this example, which typically do not generate component elements corresponding to "SoC board") do not increase their baseline object identity set due to objectification; otherwise, the target projection will experience a false failure due to the lack of corresponding model elements. Migration is implemented by appending records; the original attribute items and their historical versions are not deleted, thus maintaining the continuity of the evidence chain before and after objectification. Objectification does not change the identifier of the corresponding model element in the generated domain model; it only changes the projection relationship record from "pointing to..." U1. A certain attribute item is changed from "pointing to the created unified semantic object". The objectification event of this attribute item is also written into the evolution field as a structural change event of the unified semantic projection range, and triggers the update of the projection dependency index, the determination of the scope of influence, and the incremental reprojection of the affected target projection: before the incremental reprojection is completed and the semantic conservation quantity comparison is passed, the affected formal domain models temporarily lose their synchronization qualification. In this example, the 3D structure projection already has a model element corresponding to "SoC board", so this projection only updates the object pointed to by its projection relationship record without adding a new model element; if the hardware architecture projection declares the generation of a board element corresponding to the object but its existing model does not yet have this element, then it generates a new model element and an associated reconnection patch; the UML software architecture projection does not declare the generation of a board element corresponding to "SoC board". The component elements corresponding to the "board" are not affected by this objectification, and their eligibility for synchronization in the formal domain model is not revoked. Without this processing, a situation will arise where the object identity set of the baseline semantic conservation quantity has been included in the new object, but the corresponding target domain model does not yet have a corresponding model element. The next conservation quantity comparison will fail due to the missing identity component. In this embodiment, the 3D structure projection declares the object identity and key parameters as mandatory conservation components, and the topological relationship structure and interface signature as optional conservation components; therefore, its model element count is greater than... V The number of unified semantic objects in 1 does not cause semantic consistency check failure.

[0120] As an example of a specific implementation of the projection operator, the projection operator registered in the System Logic Modeling Domain (SysML) in this embodiment... P SysML Execution follows the following mapping relationships and creation order. The mapping relationship is: Uniform Semantic Object. U The converged concept classes are assumed to be mapped to modules (Blocks), with object identifiers written into the module's unique identifier field and concept names written into the module name; Att U The converged scalar properties in the module are mapped to the module's value properties, and their units of measurement are mapped to the units of the value type. The expected value distribution is written to the default value; Cap U The behavioral capabilities in the module are mapped to the operations it possesses and the transition triggering events of the state machine; Rel U The composition class relationships are mapped to Composite Association and PartProperty, the connection class relationships are mapped to Port and Connector, and the dependency class relationships are mapped to Dependency; UThe constraint mapping in the model is a constraint block and its binding connector. Hard constraints generate an additional requirement and establish a satisfaction relationship. The mapping condition is: only those that meet the convergence criterion of equation (6) and have a confidence level not lower than a certain threshold are satisfied. i v det Semantic information items participate in the mapping; information items in multiple hypothesis states do not generate model elements. The creation order is as follows: first create modules, then value attributes and ports, then component attributes and connectors, then constraint blocks and binding connectors, and finally the state machine and its transitions. The creation of subsequent elements depends on the identifiers of preceding elements, so the order cannot be reversed. An example of a one-to-many mapping is: U The single attribute item "Number of Cameras = 8" generates 8 ports of the same type. The projection relationship record records the correspondence between this attribute item identifier and the 8 port identifiers in a one-to-many mapping type; an example of a many-to-one mapping is: U The "Power Consumption Limit" attribute item of 1 and the constraint object "Total power consumption not exceeding 35 W" are mapped to the same constraint block. The projection relationship record records two corresponding relationships respectively in a many-to-one mapping type. The reverse promotion rule is as follows: the addition and deletion of ports are mapped to the addition and deletion candidates of connection class relationship objects; the value attribute value change is mapped to the value candidate of the corresponding attribute item and is inversely normalized by units; the constraint block expression change is mapped to the expression change candidate of the constraint object; module renaming only generates candidates at the conceptual class assumption level without changing the object identifier. The conservation quantity extraction rule is: Ext SysML The local identifier is read from the module's unique identifier field and restored to a canonical identifier according to the identifier mapping rules and projection relationships of the SysML domain to obtain the CQ. id CQ is extracted from the directed graph formed by component attributes and connectors. top CQ is obtained by extracting and normalizing the value attributes and their units. par Extract CQ from port name, direction, data type, dimension and protocol configuration int Extracting CQ from constraint block expressions conThe extraction process re-parses the generated SysML model file, without reading any intermediate data structures during the projection execution process. Operators registered in the geometric domain (CAD) and software domain (UML) are implemented according to the same paradigm, differing only in the target metamodel fields and creation order: CAD projection operators map geometric envelopes to sketches and extrusion features, and positioning datums to datum planes and mounting hole features, creating them in the order of "datum plane—main feature—additional feature—assembly constraint"; UML projection operators map behavioral capabilities to components and interfaces, and interface signatures to operation signatures, creating them in the order of "component—interface—implementation relationship—dependency relationship—timing parameter".

[0121] Step S4 (Conservation Quantity Verification): A7 From V 1. Calculate the baseline semantic conservation quantity CQ ( V 1): Identity identifier CQ id Topology CQ top (8-channel camera connection, 2-channel CAN-FD connection), key parameter CQ par (35 W, 33 ms, 1 s, all normalized to SI units), Interface Signature CQ int (CanFdDriver::send semantic summary), effective constraint set CQ con (Power consumption budget, installation tolerance). Among these, the baseline semantic conservation quantities only include the object identity and relationship that already satisfy the convergence criterion of equation (6): U 1's identity identifier is recorded in CQ id The identity of the object has not yet converged. U 2. Not included, because one end of the entity relationship object "Power Supply Relationship #1" is... U 2. This round also does not participate in projection generation; its identity and topology are not included in the baseline semantic conservation quantities. U 2. After identity convergence, all identities are included and incrementally reprojected into the corresponding domain models. Therefore, no identity exists in each domain model. U Model elements 2 and "Power Supply Relationship #1" do not constitute CQ. id With CQ top Component deviation. The 3D structure projection does not declare the topological relationship structure as a required component, and the absence of topological elements corresponding to camera connections and CAN-FD connections in the model does not constitute a verification failure. Extraction logic for conserved quantities in various fields (Ext). d After re-analyzing the generated domain model as input, it is compared according to equation (11). In this embodiment, each domain is selected. i d cons=0.05, key parameter single-item relative deviation threshold 2%. The first round of UML projection verification revealed interface signature component deviation: the component diagram omitted the DiagnosticManager diagnostic reporting interface. d UML,int =0.25 exceeds the single-item threshold i d,r cons The interface signature component is located, and a repair proposal is generated (correcting the mapping branch from behavioral capabilities to interfaces in the UML projection operator and forming a new operator configuration version). After repair, the verification is passed. All four domain models are marked as synchronized formal domain models and presented to engineers for browsing and editing via B1. As a comparative note: if the deviation is not eliminated after a maximum of 3 preset repairs, the UML model is marked as a consistency check failure state, not included in the synchronization link, and a failure message containing the deviation component location is output and presented via B1.

[0122] Step S5 (Change Evidence Feedback): The engineer changes the 2-channel CAN-FD to 3-channel in the hardware architecture diagram. The change evidence and enhancement module A3 generates a domain change evidence record (change source = model editor, operation subject = engineer identifier, model difference = addition of 1 CAN-FD transceiver and two connections, valid projection instance handle, and both the baseline semantic version identifier and the baseline model version identifier are consistent with the current one), and temporarily revokes the synchronization eligibility of the changed hardware architecture model. Whether to restore it will be determined after the change evaluation and closed-loop verification are completed; the enhancement operator registered in the hardware architecture domain... L d Based on the projection relationship record in the change promotion mode, the differential is decomposed into three atomic model changes according to the decomposition granularity of "single model element or association between single model elements" as described in Part (VII): adding a transceiver element, adding the first connection, and adding the second connection; among them, the addition of the transceiver element generates a semantic change candidate "adding the third CAN-FD bus channel", and the two connections generate new candidates for the corresponding connection class relationship objects. All three semantic change candidates are unambiguous and pass the hard constraint verification: before the change U The estimated total power consumption of device 1 is 31.8 W. The estimated power consumption increase due to the addition of the third CAN-FD transceiver and its interface circuit is 1.2 W. After the change, the estimated total power consumption is 33.0 W, which is lower than the upper limit specified by the hard constraint "total power consumption not exceeding 35 W". Therefore, all three candidates are judged to be accepted and converted into semantic evidence records. α =0.95) Entered; handle valid, A1 directly located. U1. Perform fusion on the relevant objects and the fusion result satisfies the convergence criterion of equation (6) and there are no unresolved hard constraint conflicts.

[0123] Step S5 (Affect Range and Incremental Reprojection): Influence Analysis and Incremental Reprojection Module A6 calculates the influence range Imp using equation (13) through the relation field and constraint field. ΔV 1) The number of CAN ports in the SysML internal block diagram, the channel parameters of the CanFdDriver in the UML component diagram, and the projection instances corresponding to the connector row layout in the 3D structure fall within the affected range. The power consumption constraint needs to be recalculated. A6 sends an incremental reprojection request to A7. A7 performs incremental reprojection on the three affected projections and verifies the conserved quantities one by one. Then, based on the updated... V 1. Perform closed-loop verification between the expected semantic conservation quantities in the generated hardware architecture domain and the modified hardware architecture model; the deviation should be less than [percentage missing]. i d cons The model regains its synchronized qualification; all reprojection products are marked as system source updates, no new domain change evidence records are generated, and the loop terminates. The evolution field records the cause of this change, its scope of impact, and the reprojection results.

[0124] Example of replay-based refusion: Subsequent arrival of high-resolution photographic evidence suggests that previously associated with U Evidence 1 e The "second row connector" segment in section 2 actually describes adjacent independent ECUs (objects). U 2). A1 Correct Anchoring: Move the segment from... U 1. Unassociate and reassociate with U 2. Both deactivation and reconstruction are achieved by adding related event records; subsequently, based on the immutable preservation in the evidence field... e 1 to e 5 and subsequent evidence, regarding U 1 and U 2. Perform replay-based refusion separately. The 3D structural model generated based on the semantic version before replay is temporarily deprived of synchronization eligibility and marked as an isolated version. After the completeness, incremental reprojection and conservation quantity verification are re-determined and passed, it is restored. Anchoring errors are corrected without loss, and no manual rollback is required throughout the process.

[0125] Concurrent merging example: Another engineer modifies the state machine transition conditions in the SysML diagram almost simultaneously. The two change events generate domain change evidence records respectively, which are merged sequentially according to the evidence's chronological order; because they both affect the same unified semantic projection range... V 1. After fusion, the scope of influence determination and incremental reprojection are performed only once.

[0126] Example of tracing the source: The reviewer selects channel parameter "3" of CanFdDriver in the UML component diagram, and traces its evidence chain back through projection relationship records and formal bidirectional indexes. The currently determined semantics are... U The CAN-FD channel count of 1 = 3” is formed by enhancing, accepting, and fusing the domain change evidence record (including change time, operating subject, and model difference) generated by the engineer's editing on the hardware architecture diagram in step S5. This record and its derived semantic evidence record constitute the supporting evidence set for the current conclusion. Original requirement evidence e The candidate semantic hypothesis supported by the corresponding fragment of 1 is "CAN-FD channel number = 2". This hypothesis was in a determined semantic state before this change, and it has been replaced by "channel number = 3" after this change. e 1. Following the requirement of immutability in addition, it remains in the evidence field. Its support relationship for "Number of Channels = 2" and its initial evidence support are not deleted or rewritten, but it does not constitute supporting evidence for the current conclusion. The substitution relationship, reasons for change, and scope of influence between "Number of Channels = 2" and "Number of Channels = 3" are recorded in the evolution field. Therefore, the current conclusion can be traced back step by step along the evolution field to the historical conclusion and its original supporting evidence. For comparison, the set of supporting evidence for the information item "Power Consumption Limit = 35 W" for the same object is... e The corresponding fragment of 1 and the engineer's manual review record, the candidate hypothesis that was replaced is: e 5. Supports the “40 W”; each semantic information item maintains an independent set of supporting evidence and substitution relationships. The formal bidirectional index can trace back all supporting evidence fragments and related semantic evidence records from any accepted semantic conclusion, and can also locate all conclusions it has supported and affected model elements from any semantic evidence fragment. The entire chain is auditable.

[0127] Example 2: Modeling a 3D virtual world environment based on demand input.

[0128] This embodiment uses a 3D virtual world environment as the scene. The target projection set includes three target domains: a 3D scene model (geometry, materials, lighting, and free navigation), a scene semantic graph (a logical view of objects and relationships, representing different views and abstract levels within the same modeling system as the 3D scene), and physical and interaction configurations (colliders, interactive behavior parameters). This embodiment uses the same system deployment and parameters as Embodiment 1. i anc =0.75、 i mar =0.10、 t 1 = 0.85 t 2 = 0.20 n stab =3、 c gen,d =0.50.

[0129] The user provides three semantic input data streams: a text request to "construct a scene of a seaside town at dusk: a lighthouse stands on a cape, a pier connects to the dock, three fishing boats are moored, and the scene can be freely explored"; a reference concept map (or a paused frame from a video); and subsequent audio supplementation. An external visual model analyzes the reference map to produce evidence, where each detected region constitutes an independent semantic evidence fragment: a candidate hypothesis for a tall, columnar building fragment {lighthouse: 0.70, chimney: 0.40}, a fragment of a wooden structure on the water {pier: 0.80}, and several fragments of boats {fishing boats: 0.85, sailboats: 0.35}; an external large language model analyzes the text to produce evidence of objects, relationships ("lighthouse—located on—cape", "pier—connects to—dock"), and quantity attributes (fishing boats = 3). All AI output is recorded only as semantic evidence.

[0130] Anchoring and Fusion: The image evidence fragments of the columnar building and the textual evidence fragments of the "lighthouse" are anchored to the same object via formula (4). U 11 (Spatiotemporal positioning is aligned with image and text references, and the score is anchored) z =0.83), after fusion using equation (5), U 11 The "lighthouse" hypothesis establishes confidence. With a confidence level of 0.93 and a "chimney" hypothesis of 0.07, equation (6) converges; fishing boat objects are created according to image segments. U 12 , U 13 , U 14 The text quantity attribute "3" corroborates the image detection quantity. The "located" and "connected" relationships are written into the relation field, where the "docked" relationship is materialized as an independent object carrying the soft constraint "no more than 5 meters from the edge of the dock". The unified semantic projection range for this round is denoted as... V 2.

[0131] Limited participation of historical prediction candidates: The concept map does not provide lighthouse height. The system uses a unified semantic object. U 11 Evolutionary history Based on prior knowledge of similar scenarios, a historical prediction candidate "lighthouse height approximately 30 meters" is generated. This historical prediction candidate is a restricted system candidate, lacking semantic evidence record status. It is not counted as semantic evidence in the number of evidence entries in equation (5), nor is it included in the independent evidence related group or semantic evidence source category. Instead, it serves as a candidate auxiliary factor independent of the semantic evidence likelihood term, participating in the confidence calculation of the lighthouse height candidate semantic hypothesis with restricted weights according to equation (14). In this embodiment, the preset restricted attenuation coefficient is taken as... r =0.5; Since there is no actual observational evidence of the lighthouse height from an unrestricted source, the reference weight is taken as the preset restricted baseline weight, i.e. α ref = α 0, therefore the effective weight of this historical prediction candidate satisfies α lim ≤0.5 α 0. After this historical prediction candidate was included in the confidence calculation, the confidence level of the candidate semantic hypothesis "the lighthouse is about 30 meters high" increased to 0. b v =0.62. The generation or update of the historical prediction candidates is not included in the number of updates of independent evidence related groups in the stable window of equation (6). Since all candidate support sources for the lighthouse height are restricted system candidates, there is no semantic evidence from unrestricted sources, and the height value has not yet passed independent constraint solving, simulation, or deterministic verification, even if the confidence of its candidate semantic hypothesis is improved, the semantic information item still cannot obtain a determined semantic state, nor is it included in the weighted statistics of the determined semantic information item in the target projection completeness of equation (7). When the user confirms or modifies the lighthouse height, external measurement data gives the lighthouse height, or independent geometric constraint solving verifies the height value, the system generates a semantic evidence record separately according to the corresponding user confirmation result, external measurement data, or independent verification result, and retains the historical prediction candidate identifier on which it is based in the semantic evidence record. The newly generated semantic evidence record participates in evidence fusion according to the ordinary semantic evidence rules; the original historical prediction candidates continue to be retained as candidate generation process records and are not deleted or rewritten due to the generation of new semantic evidence records.

[0132] Completeness and Draft Presentation: The scene semantic graph sub-completeness Cmp is calculated according to Equation (7). d ( V 2) = 1.0, physical configuration 0.96, 3D scene 0.68, scale-related information items (lighthouse height, pier length) have not converged. The 3D scene's Gen... d =0 and 0.68≥ c draft,dThe second branch of formula (9) is applied. A7 generates a draft state domain model of the 3D scene. After passing the structural validity check (complete scene graph nodes and parent-child references, no isolated nodes) and the non-exempt hard constraint check (no geometric interference), the draft model is presented in a semi-transparent style by B1. Users can roam and preview, but the interface is marked "Draft: Scale to be determined". At the same time, A5 selects the acquisition action "ask the user about the height of the lighthouse and the length of the pier" according to formula (10) (gain 2.1 bits, low cost, better than "request a picture with scale reference" 1.6 bits, medium cost). The user directly drags and adjusts the lighthouse to the ideal scale in the draft scene, and then confirms by voice "lighthouse height 35 meters, pier 60 meters". This drag-and-drop editing is located by the lifting operator of the 3D scene domain according to the draft projection relationship record (draft model version draft-1) established when the draft was generated. U 11 The height attribute item generates a draft change evidence record carrying the draft projection instance handle, the baseline draft model version identifier draft-1, and the task version identifier. Its preliminary validity check is based on the current draft model version, not the formal domain model version (at this time, there is no formal domain model in the 3D scene domain). Both the draft change and the voice response are entered into the field as evidence (user observation evidence). α =0.9), overwhelming historical prediction candidates, the scale information term satisfies the convergence criterion, Cmp d ( V 2) Upgraded to 1.0 and Gen d Converted to 1, A7 from the updated V 2. Re-execute the formal projection (without transferring draft content). In this example, the scene semantic graph and the physical configuration are generated in the same round as the formal projection of the 3D scene. Therefore, there is no existing formal domain model that needs to be updated separately. If the scene semantic graph has previously obtained synchronization eligibility, the scale class introduced by the draft editor has accepted the semantic change and also generated a unified semantic state change event. The synchronization eligibility of the formal domain model of the scene semantic graph is temporarily revoked, and then incremental reprojection is performed on it and restored after verification of the conservation quantity. The geometric content of the draft scene itself is not applied to the scene semantic graph.

[0133] Conservation Quantity Verification: The three domain models are processed through their respective conservation quantity extraction logics (Ext). d After re-parsing using the generated model as input, it is compared with the baseline semantic conservation quantity CQ. V2) Compare according to formula (11): object identity (lighthouse, pier, fishing boat ×3), topology (located in relation, connected in relation, docked in relation), key parameters (lighthouse 35 meters, pier 60 meters, both normalized), constraint set (roaming passage width not less than 2 meters, fishing boat distance from dock not greater than 5 meters). The first round of physical configuration projection found that "pier collision body width 1.8 meters" violated the passage constraint, and the deviation was located to the effective constraint set component CQ. con The proposal to fix the default collider width parameter of the physical projection rules has been verified and all three models have been marked as synchronizable formal domain models and rendered.

[0134] Reverse synchronization: The user deletes a fishing boat in the 3D scene. A3 generates a domain change evidence record, and A7 temporarily revokes the synchronization eligibility of the changed 3D scene model; the 3D domain lifting operator decomposes the deletion operation into atomic model changes, and generates two mutually exclusive candidates according to the conversion rules of deletion-type atomic model changes: candidate one "sets the fishing boat object to a tombstone state" and candidate two "hides the object only in the domain model". The semantic consequences of the two are different and cannot be automatically distinguished by the model difference itself. According to the triage priority described in Part (VII), the two are set to a pending user confirmation state and a confirmation request is output; after the user confirms "permanent deletion", the confirmation result serves as a new high-reliability semantic evidence record ( α =0.98) enters the field, and the candidate meets the acceptance condition and obtains the acceptance state and re-participates in the fusion of equation (5). The confidence of its fusion result satisfies the convergence criterion of equation (6) and there is no unresolved hard constraint conflict, thus forming an accepted semantic change. U 14 The system is then set to a tombstone state. User confirmation does not directly trigger the tombstone state; instead, it undergoes the same four stages as other semantic inputs: evidence entry, obtaining acceptance status, fusion, and convergence. "Obtaining acceptance status" only indicates obtaining fusion access; "forming an accepted semantic change" is the condition for writing the determined semantic state. Subsequently, the system analyzes and locates the scene semantic graph (removing corresponding nodes and docking relationships) and physical configuration (removing colliders). A6 issues an incremental reprojection request, which A7 executes and verifies. Simultaneously, A7 updates the system based on the... V 2. Generate the expected semantic conservation quantities of the 3D scene domain, perform closed-loop verification with the modified 3D scene model, and if the deviation is within the threshold, the model is restored to synchronization eligibility; all reprojection products are marked as system source updates to prevent echoes. The scene semantic graph and the 3D scene are kept consistent through a unified semantic projection range, and there is no direct mapping between the two domain models throughout the process.

[0135] Continuous evolution: When users add "add a loop trail around the town" as additional input, the new evidence enters the same closed loop, the unified semantic projection range expands, and the projection of the three domains is updated incrementally. The modeling process is characterized by the continuous growth of the semantic field, rather than a one-time generation.

[0136] like Figure 1 As shown, the system of the present invention includes at least one processor and a memory. The computer instructions stored in the memory are implemented when executed by the processor: evidence generation module A2, semantic object association and fusion module A1, completeness determination module A5 (active completion decision is its sub-function), projection and consistency verification module A7, change evidence and enhancement module A3, impact analysis and incremental reprojection module A6; dynamic semantic state field A4 is a data structure rather than a processing module, and is read and written by the above modules through a transactional interface. The access and interaction layer B1 (input adapter group, question-and-answer confirmation channel, model presentation and editor) and the domain model B2 are the system boundary entities: users have three roles: information source, arbitrator, and change initiator. Their outputs enter the system through mutually exclusive pathways. Outputs used to determine the semantic content of objects enter the system via A2 as semantic evidence records. Editing operations on domain model elements are performed on B2 via B1, and change events issued by B2 enter the system via A3 as domain change evidence records. Control configuration inputs used to set target projection sets, thresholds, permissions, and result presentation methods do not pass through A2 and directly update the running configuration of each module and the version configuration of the current task. The four final states—formal domain model, draft domain model, incomplete prompts, and consistency check failure prompts—all reach the user via B1, and there are no silent dead zones in the system.

[0137] Data flow summary: B1 → A2 (semantic input data) → A4 evidence field (immutable write) → A1 (anchoring and fusion in fragment units, reading and writing A4) → A5 (reading converged terms in A4 to calculate completeness; if insufficient, data is collected and fed back to A2 via B1; if satisfied, A7 is triggered) → A7 (unique projection execution point, forward projection, draft generation, conservation quantity verification, and creation of projection relationship records; the product is sent to B2 and presented via B1) → B2 change event → A3 (evidence enhancement and writing to A4) → A1 fusion → A6 (reading A4 to determine the scope of influence, sending an incremental reprojection request to A7) → A7 (execution and verification, system source marking). Each module can be deployed on the same electronic device or distributed (e.g., a cloud-edge collaborative model where A2 parsing is performed at the edge and fusion and projection are performed in the cloud). The computer program product described in the fourth aspect of this invention can be provided as a downloadable software installation package, a standalone executable program, a domain operator package provided as a plug-in (containing the registration implementation of projection operators, lifting operators, and conservation quantity extraction logic for a specific target domain), a container image, a virtual machine image, a cloud deployment program, or a combination of the above forms. After being downloaded, installed, or deployed to an electronic device with a processor and memory, the processor executes the computer program contained therein to implement the aforementioned steps S1 to S5 and each preferred embodiment. The functional module division implemented is the same as that of the aforementioned system implementation. The computer program product can be a complete program containing all functional modules, or it can be an incremental program that only adds or updates projection operators, lifting operators, and conservation quantity extraction logic for a specific target domain. After the latter is deployed, as described in Part (VI), it does not change the core data structure of the dynamic semantic state field and the unified semantic object, nor does it change the projection of other target domains. It only marks the projection relationship records affected by the version change as to be reconstructed.

[0138] For persistence, entity and relation fields can use graph storage, evidence fields can use append-only log storage, and evolution fields can use event-based storage. Alternatively, a combination of relational databases, key-value stores, or object stores can be used, but the two key attributes of logical immutability and traceability of conclusions must be maintained. Physical layer compression and archiving do not violate logical immutability. Furthermore, to make the aforementioned logical immutability constraints verifiable attributes at the physical storage level, each record in the evidence field can be appended with a content digest, a previous record digest, a writing subject identifier, and a writing time. A digital signature or message authentication code can also be appended to these, thereby performing integrity checks on the record sequence. Records that fail the check are marked as integrity anomalies and a prompt message is output; they do not participate in subsequent evidence fusion, nor are they used as input for replay-based refusion. This integrity mechanism is optional; its absence does not affect the implementation of the other mechanisms of this invention. However, when this mechanism is absent, the immutability referred to in this specification only means that the system does not modify or delete existing records at the logical semantic level.

[0139] Regarding algorithm replacement: the probabilistic inference model in equation (5) is preferably implemented using a factor graph, but it can also be replaced by equivalent methods such as Bayesian networks, Markov random fields, Dempster-Shafer evidence combinations, or neural network output and iteratively updated confidence, only needing to maintain the interface convention of "evidence support input → confidence update → convergence determination"; the similarity sim in equation (4) can be replaced by inner product, negative distance, or learning-type metric; the uncertainty Unc in equation (10) can be replaced by Shannon entropy or other uncertainty metrics such as variance; the layer Laplace extension in equation (12) is an optional implementation. (The thresholds and coefficients are not directly related to the algorithm replacement.) i anc , i mar , t 1. t 2. n stab , i conf , c gen,d , c draft,d , i v det , i d cons , i d,r cons , i glob , r All values ​​can be set by the control configuration input according to the accuracy requirements of the field. The values ​​used in the embodiment are only examples.

[0140] Complexity and Runtime Performance: In implementations employing sparse relational structures, local incremental updates on factor graphs, projection-dependent indexes, and asynchronous task queues, this approach limits computation and reprojection to the affected semantic elements and model elements, thus avoiding full recalculation and full model regeneration: evidence fusion only reaches the estimated variables affected by new evidence; completeness calculation is a weighted statistical representation of pre-declared semantic information items; and the scope of influence calculation restricts propagation to the reachable closure Cl( ΔV Within this scope; parsing and fusion can be executed asynchronously with the interaction process, thereby reducing blocking on interaction responses and state field read / write operations. Cross-domain synchronization propagates via a star topology, requiring a minimum number of registered operator plugin groups and target domains. n DLinear dependence. The above runtime performance depends on specific implementation conditions, including the sparsity of the relational structure, the scale of the impact of a single change, the validity of the projection dependency index, whether local factor updates and asynchronous task queues are used, and whether there are cross-range global constraints that cause patches to cascade over a large area; in the case of dense relational structures or large-scale cascading constraints, the actual computational load will increase accordingly.

[0141] The target domain is not limited to the type shown in the ...

Claims

1. A cross-domain modeling method for unified semantic objects, applied to a computer modeling system, wherein the computer modeling system registers projection operators and lifting operators for at least two target domains respectively, and configures conservation quantity extraction logic for each target domain, characterized in that... The method includes: S1. Acquire input data from one or more information sources, and generate semantic evidence records based on semantic input data used to determine the semantic content of the object in the input data. The semantic evidence records include at least source information and one or more semantic evidence fragments. Each semantic evidence fragment includes location information, candidate semantic hypotheses, and initial evidence support corresponding to each candidate semantic hypothesis. Store the semantic evidence records in a dynamic semantic state field in an append-only immutable manner. The semantic input data from any information source participates in the update of the unified semantic object in the form of semantic evidence records, without directly overwriting the determined semantic state of the unified semantic object. The determined semantic state includes at least one of the following: semantic state, attributes, relationships, behavioral capabilities, and constraints that have met the convergence conditions. S2. Using semantic evidence fragments as units, associate and match each semantic evidence fragment with existing unified semantic objects in the dynamic semantic state field. These unified semantic objects are categorized into entity-type, relational-type, and constraint-type unified semantic objects based on the semantic element types they carry. Relational-type unified semantic objects carry typed relationships between two or more unified semantic objects, while constraint-type unified semantic objects carry verifiable constraint expressions acting on one or more unified semantic objects and their satisfied states. In this method, relational objects and constraint objects refer to relational-type and constraint-type unified semantic objects, respectively. The semantic evidence fragment is then associated based on the association and matching results. To any existing unified semantic object of any type, or according to the semantic element type indicated by the semantic evidence fragment, create a new entity-type unified semantic object, relation-type unified semantic object, or constraint-type unified semantic object and associate the semantic evidence fragment with the created unified semantic object. Different semantic evidence fragments in the same semantic evidence record can be associated with different unified semantic objects respectively. Merge one or more semantic evidence fragments associated with the same unified semantic object, update at least one of the semantic state, attributes, relationships, behavioral capabilities, and constraints of the unified semantic object, and retain semantic content that has not yet met the convergence conditions as a candidate semantic hypothesis state without writing it into the determined semantic state. S3. Obtain the target projection set corresponding to this round of modeling tasks. The target projection set includes one or more target projections corresponding to each of the registered target domains. Determine the unified semantic projection range corresponding to the target projection set. The unified semantic projection range includes one or more entity-type unified semantic objects, and at least one of relational unified semantic objects associated with the entity-type unified semantic objects and constraint unified semantic objects acting on the entity-type unified semantic objects. When there are no relational unified semantic objects or constraint unified semantic objects, the unified semantic projection range is composed of the one or more entity-type unified semantic objects. Determine the completeness of the unified semantic projection range relative to each target projection according to the semantic information requirements declared by each target projection. For target projections whose completeness meets the preset model generation conditions, ... The unified semantic projection range is projected to generate a corresponding domain model using the projection operator registered in the target domain corresponding to the target projection. A projection relationship record is established and stored between the semantic elements in the unified semantic projection range and the model elements in the generated domain model. The semantic elements refer to entity-type unified semantic objects, relation-type unified semantic objects, and constraint-type unified semantic objects in the unified semantic projection range. The domain models corresponding to at least two different target domains are generated by executing this step once or multiple times. After the domain models of at least two different target domains obtain synchronization eligibility, they are included in cross-domain change propagation. For target projections whose completeness does not meet the preset model generation conditions, the formal domain model generation for that target projection in this round is stopped, or at least one of the following is entered: active completion process or draft state domain model generation process. S4. Each target projection declares a set of mandatory semantic conservation components, a set of optional semantic conservation components, a deviation threshold corresponding to each semantic conservation component, and a normalization rule version corresponding to its target model type. The set of mandatory semantic conservation components includes at least one of the following: object identity identifier, topological relationship structure, key parameters, interface signature, and effective constraint set. A baseline semantic conservation quantity corresponding to the semantic conservation components declared by each target projection is obtained from the unified semantic projection range. Through conservation quantity extraction logic set corresponding to each target domain, using at least one of the persistent model content, executable structure, or parsable model structure of the generated domain model as input, and following a model parsing path independent of the corresponding projection operator, normalized semantic conservation quantities are obtained from each domain model through parsing. The constant quantity extraction logic does not reuse the semantic conservation quantity values ​​directly generated by the projection operator during the generation of the corresponding domain model or the intermediate results used to construct the semantic conservation quantity values. The conservation quantity extraction logic and the projection operator can share the domain meta-model and normalized base library of the corresponding target domain. The mandatory semantic conservation quantity components obtained from parsing each domain model and the actual optional semantic conservation quantity components are compared with the corresponding components in the benchmark semantic conservation quantity. The absence of components that are not declared as mandatory semantic conservation quantity components by the corresponding target projection does not constitute a semantic consistency check failure. Based on the consistency deviation obtained from the comparison, it is determined whether the corresponding domain model meets the semantic consistency condition, and only the domain model that meets the semantic consistency condition is marked as a synchronizeable formal domain model. S5. For model updates actually applied to the domain model by the computer modeling system through incremental reprojection, an additional system source update identifier is added. This system source update identifier, along with the corresponding unified semantic state version identifier, the updated target model version identifier, and the update operation record, is stored in the dynamic semantic state field. Only when the system source update identifier and its associated information carried by the model update event match the corresponding record in the dynamic semantic state field is it determined to be a valid system source update identifier. When a model update event carrying a valid system source update identifier is received, no domain change evidence record is generated for that model update event, and no boosting operator processing or evidence fusion is performed. If the system source update identifier carried by the model update event cannot match the corresponding record in the dynamic semantic state field, it is treated as a change event without a valid system source update identifier. The expected model or expected semantic conservation quantity used for closed-loop verification of the source domain model semantic conservation quantity is only used as a verification product within the computer modeling system, not written into the corresponding formal domain model, not entered into the domain model change event listening channel, and not used as a system source update. When a model update event without a valid system source update identifier is received from the first formal domain model that can be synchronized in the first target domain, it is considered a valid system source update event. When a system source update identifier change event occurs, an appended immutable domain change evidence record is generated based on the change event, and the synchronization qualification of the first formal domain model after the change is temporarily revoked, so that it does not participate in cross-domain change propagation before the change processing is completed; through the promotion operator registered in the first target domain, based on the domain change evidence record, the model difference before and after the change of the first formal domain model, and the projection relationship record, the model difference is decomposed into one or more atomic model changes, and semantic change candidates involving one or more semantic elements in the unified semantic projection range are generated respectively; state determination and corresponding processing are performed on each semantic change candidate, wherein the semantic change candidate that obtains the acceptance state through state determination is associated with the domain change evidence record as a candidate semantic interpretation and participates in evidence fusion. Only when its fusion result satisfies the convergence condition and there is no unresolved hard constraint conflict that forms an accepted semantic change, the fusion result is used to update the determined semantic state of the unified semantic object; semantic change candidates that do not obtain the acceptance state, and semantic change candidates that obtain the acceptance state but do not form an accepted semantic change after fusion, do not update the determined semantic state;For semantic change candidates that have been accepted through evidence fusion, before determining the scope of influence of the change event and performing the semantic conservation closed-loop verification of the first formal domain model, based on the atomic model change type corresponding to the semantic change candidate, the unified semantic object localization result obtained directly from the valid projection instance handle and projection relationship record, or obtained through association matching when the projection instance handle is missing, invalid, or ambiguous, and the updated unified semantic projection range, the formal projection relationship records between the model elements, associations, or attribute items in the first formal domain model and the corresponding unified semantic object or the corresponding unified semantic object attribute item are established, updated, terminated, or invalidated. The atomic model is modified by adding a new model element, a new association, or a new constraint. A projection instance handle is generated for the corresponding model element, association, or constraint, and a formal projection relationship record is established, containing the corresponding unified semantic object identifier or attribute item identifier, the local identifier in the first formal domain model, the mapping type, the source semantic version, the current version of the first formal domain model, the projection operator identifier and version, and the promotion operator identifier and version. When the atomic model is modified by updating an existing model element or attribute item, at least one of the source semantic version, the target model version, and the mapped attribute item identifier in the corresponding formal projection relationship record is updated, while retaining the historical version before the update. When an atomic model is changed to an association reconnection, the endpoint references of the corresponding relational unified semantic object and the formal projection relationship record between the relational unified semantic object and the associated elements of the source model are updated. When the atomic model is changed to the deletion of model elements, associations, or constraints, depending on whether the accepted semantic change represents the semantic deletion of the unified semantic object or only the removal of the projection product in the first target domain, the corresponding formal projection relationship record is marked as terminated due to semantic deletion or invalidated due to the removal of the source projection, and its historical record is retained. The establishment, update, termination, or invalidation markers of the above-mentioned formal projection relationship records are recorded in the dynamic semantic state field by appending projection relationship change events, without deletion or... The original projection relationship record is overwritten; only after the aforementioned processing is completed on the source-side formal projection relationship record related to the accepted semantic change, the corresponding model element, association, or attribute item is entered as the current valid projection instance into the current valid semantic view of the first formal domain model; when the formal projection relationship record processing fails, the accepted semantic change that has already been formed is not rolled back, but the corresponding projection relationship record is marked as pending reconstruction, so that the first formal domain model retains its synchronization qualification and is revoked, and its synchronization qualification is restored only after the reconstruction of the projection relationship record and the semantic conservation quantity closed-loop verification are completed; semantic change candidates that have not formed accepted semantic changes do not establish or activate new formal projection relationship records;For semantic change candidates that have not yet formed accepted semantic changes, at least one of the following is performed on the model elements, relationships, or attribute items in the first formal domain model corresponding to the semantic change candidate, based on their status: partial rollback, isolation marking, waiting for user confirmation, waiting for supplementary evidence, conflict resolution, or rebasing. The isolation marking is used to prevent the corresponding model elements, relationships, or attribute items that carry the semantic content of the object from entering the current valid semantic view of the first formal domain model. Model elements, relationships, and attribute items in the isolation marking state are retained in the isolation version and do not participate in semantic conservation quantity extraction or cross-domain changes as currently valid projection instances. The model difference baseline for propagation and subsequent formal domain model versions; the currently valid semantic view consists of model elements, relationships, attributes, and constraints that are not in an isolated state, exist as currently valid projection instances, and carry object semantic content; canvas coordinates, primitive styles, view collapse states, annotations, and editor auxiliary elements that do not carry object semantic content are not model semantic content that needs to be supported by the currently determined semantic state; among them, semantic change candidates determined to be in a rejected state are considered to have completed processing only after their corresponding model elements, relationships, or attribute items have undergone partial rollback or isolation marking; semantic change candidates determined to be in a candidate retention state Semantic change candidates that have obtained an accepted status but have not yet formed accepted semantic changes after fusion are considered to have completed processing only after their candidate semantic assumptions are retained in the hypothesis field of the dynamic semantic state field and their corresponding model elements, relationships, or attribute items have undergone partial rollback or isolation marking. Semantic change candidates in the pending user confirmation state or conflict state will undergo a new state determination after receiving user confirmation or modification results, supplementary evidence, or conflict resolution. Only after the new determination will accepted semantic changes be formed, or they will be determined to be in a rejected state and the corresponding model elements, relationships, or attribute items will undergo partial rollback or isolation marking, or they will be confirmed as rejected. A semantic change candidate is considered complete only when it is designated as a candidate retention state and the corresponding model element, relationship, or attribute item has undergone partial rollback or isolation marking. Semantic change candidates that did not enter the state determination due to integrity verification anomalies or inconsistencies between the baseline semantic version and the baseline model version are considered complete only when the state determination is re-executed after correction, rebasing, or re-promotion and any of the aforementioned completed actions are achieved, or when the corresponding model element, relationship, or attribute item has undergone partial rollback or isolation marking. The scope of influence of the change event is determined based on the relationships and constraints in the updated unified semantic projection range only when the accepted semantic change is formed.When one or more second target domain projections are affected within the scope of influence, the synchronization eligibility of the corresponding formal domain models of each affected second target domain is temporarily revoked. Then, based on the updated unified semantic projection scope and the projection operators registered by each second target domain, incremental reprojection is performed on each affected second target domain projection. Synchronization eligibility is restored only after the corresponding incremental model patch is successfully applied and the resulting domain model passes the semantic conservation comparison. The determination of the scope of influence and incremental reprojection can be triggered not only by the change event of the first formal domain model but also by the unified semantic state change event: when the semantic evidence record generated from semantic input data from any information source is fused... When a semantic change is accepted, causing a change in the defined semantic content within the unified semantic projection scope, and this change is not caused by a change event in a synchronizable formal domain model, a unified semantic state change event is generated. The scope of influence of this event is determined based on the relationships and constraints within the updated unified semantic projection scope and the projection relationship record. For each target projection affected within the scope, the synchronizable eligibility of its corresponding formal domain model is temporarily revoked. Incremental reprojection is then performed, and the synchronizable eligibility of each model is restored only after the incremental reprojection is completed and the resulting domain model passes the semantic conservation comparison. In this case, the semantic conservation closed-loop verification of the source domain model is not performed. This is done after all semantic change candidates related to this change have been completed. After the aforementioned processing, regardless of whether an accepted semantic change has occurred, a verification expectation model or expected semantic conservation quantity for the first target domain is generated based on the current unified semantic projection range. When an accepted semantic change has occurred, the current unified semantic projection range is the unified semantic projection range updated based on the accepted semantic change; when no accepted semantic change has occurred, the current unified semantic projection range retains the determined semantic content prior to the occurrence of this change event. A semantic conservation quantity closed-loop verification is performed using the current valid semantic view of the first formal domain model as the verification object. Verification is only performed when the current valid semantic view does not contain semantic content related to this change and carrying an object whose semantic content has not been affected by the current determined semantic state and corresponding formal projection. The system records model elements, relationships, or attribute items that support and have not undergone partial rollback or isolation marking, and whose consistency deviation obtained from the semantic conservation quantity closed-loop verification is within the corresponding threshold. In such cases, the synchronizeability of the modified first formal domain model is restored. If the consistency deviation exceeds the corresponding threshold, a repair proposal is generated or a manual confirmation request is output. Synchronization eligibility is only restored after the corresponding actions are re-executed and the semantic conservation quantity closed-loop verification passes. The first target domain differs from the second target domain; cross-domain change propagation is performed via the unified semantic projection range and projection and promotion operators in the dynamic semantic state field, without directly applying changes from the first formal domain model to the domain model of the second target domain.

2. The method according to claim 1, characterized in that, The dynamic semantic state field is a global data structure that carries all semantic content of unified semantic modeling. The unified semantic object is the basic unit that carries semantic content in the dynamic semantic state field. The unified semantic object includes a confidence function, a typed relational structure, attribute functions that change with time and context, behavioral capabilities, a set of constraints, evidence tracing references, evolutionary history, and projection relations. The confidence function is used to give the current confidence of any candidate semantic hypothesis of the unified semantic object. The candidate semantic hypotheses include concept class hypotheses, attribute value hypotheses, and relation hypotheses. The dynamic semantic state field includes: (1) an entity field for storing entity-type unified semantic objects and their semantic states; (2) a relation field for storing relation-type unified semantic objects and the relationships between unified semantic objects; (3) a constraint field for storing constraint-type unified semantic objects, constraints and their satisfaction states; (4) an evidence field for storing semantic evidence records in an append-only immutable manner; (5) an evolution field for storing object versions, reasons for changes and scope of influence; and (6) a hypothesis field for storing undetermined candidate semantic hypotheses and their confidence levels. The relations in the relation structure and the constraints in the constraint set can exist as inline items held by the unified semantic object itself, or they can be materialized into relation-type unified semantic objects and constraint-type unified semantic objects, respectively. After materialization, the relation or constraint itself has the aforementioned confidence function, relation structure, attribute function, behavioral capabilities, constraint set, evidence tracing reference, evolutionary history and projection relation components. The corresponding items in the system structure or constraint set are converted into references to the materialized unified semantic objects; entity-type unified semantic objects are stored in the entity field, relation-type unified semantic objects are stored in the relation field, and constraint-type unified semantic objects are stored in the constraint field; wherein, any relation or constraint that needs to generate independent model elements in the target domain, or needs to participate in independent reverse promotion and independent influence scope determination, is materialized into the corresponding relation-type unified semantic object or constraint-type unified semantic object before performing formal projection, thereby enabling the projection relation record to be established with the object identifier of the relation-type or constraint-type unified semantic object as the mapping unit, and enabling atomic model changes for the model element to be located to the relation-type or constraint-type unified semantic object; inline relations and inline constraints that have not been materialized do not generate independent model elements separately, their manifestation in the target domain is projected along with their respective unified semantic objects, and references to the relations and constraints on which the mapping depends are recorded in the projection relation record.

3. The method according to claim 1, characterized in that, The association matching in step S2 is performed through evidence object anchoring. Evidence object anchoring is based on semantic evidence fragments and includes: calculating at least one of the following: spatiotemporal location overlap, semantic feature similarity, and relational context consistency between the semantic evidence fragment and an existing unified semantic object; calculating an anchoring score based on the obtained scoring components; when some scoring components are missing, ignoring the missing components, setting a preset default value for the missing components, or re-normalizing the weights of the remaining scoring components before calculation; when the highest anchoring score is not lower than a preset association threshold, and the difference between the highest anchoring score and the second-highest anchoring score is not lower than a preset anchoring margin threshold, the semantic evidence fragment is determined to be associated with the unified semantic object corresponding to the highest anchoring score; when there is only one unified semantic object with an anchoring score not lower than the preset association threshold in the dynamic semantic state field, and no second-highest anchoring score exists, the anchoring margin condition is waived, or the second-highest anchoring score is taken as a preset baseline value before calculating the difference; when the highest anchoring score is not lower than the preset association threshold... However, when the difference is lower than the anchoring margin threshold, the semantic evidence fragment is retained as a candidate association, awaiting subsequent semantic evidence records or user confirmation before final determination. The result of user confirmation is written into the dynamic semantic state field as a new semantic evidence record or an appended association confirmation event. After the association state is updated, a definite association is formed, and historical candidate association records are not deleted or rewritten. When there is no existing unified semantic object, or the highest anchoring score is lower than the preset association threshold, a new unified semantic object is created and the semantic evidence fragment is associated with the created unified semantic object. The semantic feature similarity is calculated by mapping the semantic evidence fragment and the unified semantic object to a shared semantic feature space. The relational context consistency is calculated based on the degree of matching between the relational information indicated by the semantic evidence fragment and the relational structure established by prior evidence fusion in the dynamic semantic state field. The anchoring result carries the anchoring confidence and can be corrected based on subsequent semantic evidence records.

4. The method according to claim 1, characterized in that, The evidence fusion in step S2 includes: using at least one of the semantic state, attributes, relationships, behavioral capabilities, and constraints of a unified semantic object as variables to be estimated, and using the evidence support of semantic evidence fragments, pre-set prior knowledge, and constraints as influencing factors to establish a probabilistic inference model; when a new semantic evidence record arrives, only the variables to be estimated affected by the new semantic evidence record are incrementally updated to update the confidence of the candidate semantic hypothesis; for the same mutually exclusive candidate semantic hypothesis group corresponding to the semantic information item to be determined, the mutually exclusive candidate semantic hypothesis group is determined to converge when the following conditions are met simultaneously: (1) the confidence of the candidate semantic hypothesis with the highest confidence is not lower than the first threshold; (2) the difference in confidence between the candidate semantic hypothesis with the highest confidence and the mutually exclusive candidate semantic hypothesis with the second highest confidence is not lower than the second threshold; (3) in the evidence update of a preset number of consecutive times, the candidate semantic hypothesis with the highest confidence remains unchanged; when there is only one mutually exclusive candidate semantic hypothesis group When a candidate is valid, the confidence difference condition is waived; semantic content that does not meet the convergence condition continues in a multi-hypothesis state; when the conflict degree between semantic evidence records exceeds the conflict threshold, weighted arbitration is performed according to the reliability of the evidence source, and conflicts that cannot be automatically resolved generate manual review items, and the manual review results are used as semantic evidence records to participate in fusion again; wherein, the behavioral capabilities and constraints can either directly participate in fusion as variables to be estimated, or be determined according to the fused semantic states, attributes and relationships according to preset derivation rules; each semantic evidence record also carries an original source identifier and a derivation chain identifier, and multiple semantic evidence fragments from the same original input or the same derivation chain are grouped into the same evidence-related group, and semantic evidence fragments within the same evidence-related group are not repeatedly counted as independent evidence in the confidence update, but are weighted by merging according to a preset discount coefficient or only the one with the highest source reliability is weighted; the consecutive preset number of evidence updates are calculated according to the number of updates of independent evidence-related groups.

5. The method according to claim 1, characterized in that: Each target projection in the target projection set declares the semantic information items required to generate the corresponding domain model, the weights corresponding to each semantic information item, and a determined threshold. Each weight is positive so that the sum of the weights in the completeness calculation is greater than zero. The semantic information items are categorized into mandatory semantic information items, optional semantic information items, and hard-constrained associated semantic information items. The completeness determination in step S3 includes: for each target projection, calculating the sub-completeness of the unified semantic projection range relative to the target projection based on the weighted proportion of semantic information items that have met the convergence condition and reached the corresponding determined threshold among all semantic information items declared by the target projection; and calculating the sub-completeness of the target projection range relative to the target projection according to the principle of only counting the target projection. The same calculation method is used for the optional semantic information items declared by the projection. The weighted sub-completeness of the optional semantic information items is calculated. The numerator is the sum of the weights of the optional semantic information items that have met the convergence conditions and whose confidence has reached the corresponding determined threshold. The denominator is the sum of the weights of all optional semantic information items declared by the target projection. When the following three conditions are met at the same time, the target projection is determined to meet the preset model generation conditions: (1) All the mandatory semantic information items declared by the target projection meet the convergence conditions and whose confidence has reached the corresponding determined threshold; (2) All the hard constraint related semantic information items declared by the target projection meet the convergence conditions and whose confidence has reached the corresponding determined threshold, so that the relevant hard constraints have the conditions for evaluation, among which the insurmountable The deterministic evaluation result of the hard constraint exemption is satisfied, the deterministic evaluation result of the hard constraint exemption is satisfied or has been effectively exempted according to the preset exemption rule and the exemption rule identifier, exemption subject and exemption reason are recorded in the dynamic semantic state field to store the evolution field of object version, change reason and impact scope; (3) the weighted sub-completeness of the optional semantic information item is not lower than the generation threshold corresponding to the target projection. When the set of optional semantic information items declared by the target projection is empty, the weighted sub-completeness of the optional semantic information item is one, and this condition is considered to be satisfied; when each target projection in the target projection set is required to be atomically generated at the same time, only in the target projection set Atomization is performed and generated simultaneously only when each target projection satisfies its own preset model generation conditions. The minimum value of each sub-completeness or the aggregate value obtained according to the preset aggregation rules is used as a statistical and presentation indicator, and does not replace the generation gating on a per-target projection basis. When a target projection does not meet the preset model generation conditions, it is handled in the following two categories according to the reasons: First, the reason of insufficient information, that is, there are semantic information items that have not yet met the convergence conditions, have not yet reached the corresponding determined threshold, or have unresolved conflicts, or the relevant hard constraints do not yet have the conditions for evaluation due to the lack of associated semantic information items. At this time, the corresponding semantic information items to be completed are determined and the active completion process is entered.Secondly, the constraint violation is determined by the reason, i.e., the relevant hard constraint has the conditions for evaluation, but after deterministic evaluation, its satisfaction state is not satisfied. When the non-exempt hard constraint is not satisfied after deterministic evaluation, the generation of the formal domain model for the target projection in this round is stopped. The constraint violation prompt information indicating the violated constraint, its evaluation basis, and the semantic information items involved is output to be presented to the user. It can also generate correction suggestions for the values ​​of the relevant semantic information items. The correction suggestions are output as correction candidates or interactive suggestions and are not directly used as semantic evidence in evidence fusion. New semantic evidence records are formed and participate in evidence fusion only after one of the following situations occurs: user selection. The following conditions must be met for a correction suggestion to be valid:

1. Modify or confirm the correction suggestion; 2. Verify the correction suggestion through independent constraint solving, simulation, or deterministic verification; 3. Provide new input data from external files, sensors, or other information sources to support the value of the correction suggestion; 4. If none of the above conditions apply, the correction suggestion will not form a semantic evidence record, but will only be retained as a candidate correction suggestion; 5. The candidate correction suggestion is a restricted system candidate generated internally by the system and has not yet obtained independent external semantic evidence support or user confirmation. It does not have the status of a semantic evidence record, and is not included in the number of evidence items, independent evidence related groups, or semantic evidence source categories in evidence fusion, but is considered as an independent semantic evidence likelihood. The candidate auxiliary factors of the item participate in the confidence calculation of the corresponding candidate semantic hypothesis with preset limited weights; for semantically equivalent correction suggestion candidates generated by the same baseline semantic version, the same generator version, and the same target semantic information item, replacement, merging, or discount weighting are performed, and they are not accumulated repeatedly as mutually independent candidates; the generation or update of the correction suggestion candidate is not included in the number of updates of the independent evidence related group used in the convergence determination, and it shall not be used as the sole support source to enable the corresponding semantic information item to obtain a determined semantic state, nor shall it trigger the formal generation, synchronization, or incremental reprojection of the domain model alone; the correction suggestion candidate is selected, modified, or After confirmation, independent constraint solving, simulation, or deterministic verification, or obtaining supporting data from external files, sensors, or other information sources, a new semantic evidence record is generated and participates in evidence fusion according to ordinary semantic evidence rules; when an exemptable hard constraint is not satisfied by deterministic evaluation and has not yet obtained a valid exemption, the value correction process or the exemption approval process according to the preset exemption rules is entered; when there is no semantic information item to be supplemented due to a constraint violation, no collection action is performed; the active completion process includes: selecting the collection action with the largest ratio of expected uncertainty reduction to collection cost from the candidate collection action set;The candidate collection action set includes at least one of the following: posing a question to the user regarding the semantic information item to be completed, requesting supplementary input data for a specified modality, reading a specified file, invoking sensors, and invoking an external artificial intelligence model. Each candidate collection action has a collection cost greater than zero. Collection actions without invoking permissions, currently unavailable, or for which effective collection cost cannot be obtained are not included in the candidate collection action set. When the ratios of two or more collection actions are equal, the selected collection action is determined according to source reliability, expected response latency, or preset priority. The supplementary input data obtained by the collection action is reused to generate semantic evidence records. When a preset active completion termination condition is met, the current round of active completion is stopped, the unified semantic projection range is marked as incomplete relative to the target projection, and an incomplete prompt indicating the semantic information item to be completed is output to the user.

6. The method according to claim 1, characterized in that, In step S4: After the object identity identifier maps the local identifier of each domain model element to the canonical object identifier of the corresponding unified semantic object through the projection relationship record, it participates in the consistency comparison in the canonical identity space. When a unified semantic object corresponds to multiple model elements, a canonical identity set is formed based on the mapping type in the projection relationship record and the mapped attribute item identifier, and then it participates in the comparison. Topological relationships are represented uniformly before being compared for consistency. Key parameters are normalized in terms of units of measurement before being compared for consistency. Interface signatures are normalized in terms of interface name, direction, data type, dimension, and protocol before being compared for consistency. Effective constraint sets are normalized in terms of expression before being compared for consistency. Normalized consistency deviations are calculated for each type of semantic conservation quantity, and the total consistency deviation is obtained based on the deviation weights of the corresponding domains. When a single consistency deviation or the total consistency deviation exceeds the corresponding threshold, the semantic conservation quantity component with the deviation is located, and a repair proposal is generated for at least one of the following: unified semantic projection range, projection operator or configuration rules for the projection operator, and domain model. After the repair is executed, the consistency is re-evaluated. A comparison of semantically conserved quantities is performed. Specifically, repair proposals for a unified semantic projection range do not directly rewrite the unified semantic object. Furthermore, the suggested semantic content values ​​are submitted for evidence fusion only after user selection or confirmation, independent constraint solving or deterministic verification, or acquisition of independent source input data outside the repair proposal itself. Without any of the above conditions, the repair proposal does not form a semantic evidence record but is only retained as a repair proposal candidate. These repair proposal candidates are limited system candidates generated internally by the system and have not yet obtained independent external semantic evidence support or user confirmation. They do not have the status of a semantic evidence record and are not counted in the evidence count during evidence fusion. Independent evidence-related groups or semantic evidence source categories, as candidate auxiliary factors independent of the semantic evidence likelihood term, participate in the confidence calculation of the corresponding candidate semantic hypothesis with preset limited weights; for semantically equivalent repair proposal candidates generated by the same baseline semantic version, the same generator version, and the same target semantic information item, replacement, merging, or discount weighting are performed, and they are not accumulated repeatedly as mutually independent candidates; the generation or update of the repair proposal candidates is not included in the number of updates of independent evidence-related groups used in the convergence determination, and they shall not be used as the sole support source to enable the corresponding semantic information item to obtain a determined semantic state, nor shall they trigger the formal generation, synchronization, or incremental reprojection of the domain model alone; the repair proposal candidates are... After the user selects or confirms, and after independent constraint solving or deterministic verification, or after obtaining input data from an independent source other than the repair proposal, a new semantic evidence record is generated and participates in evidence fusion according to ordinary semantic evidence rules; after the execution of the repair proposal for the projection operator or its configuration rules, a new projection operator version or configuration version is formed. The projection relationship record that depends on the old version is marked as to be rebuilt and re-established when the projection is re-executed according to the new version. The domain model obtained by re-projection must be re-executed for semantic conservation comparison before it can be synchronized; the repair proposal for the domain model is limited to repairs that do not change the serialization format, reference integrity and naming convention of the semantic content of the domain model.When the preset repair termination conditions are met, the repair process stops, the corresponding domain model is marked as having failed consistency verification, it is not included in the automatic cross-domain synchronization link, and a verification failure message containing the located deviation component is output to the user.

7. The method according to claim 1, characterized in that, In step S5: Before performing the generation of domain change evidence record based on the change event and the temporary revocation of the synchronizable qualification of the first formal domain model as described in claim 1, an idempotent identifier that can uniquely characterize the change event is obtained or generated; When the change event carries a source event identifier allocated by its information source within the namespace corresponding to the information source and not reused by other change events, the idempotent identifier is generated jointly based on the source identifier of the information source, the first target domain identifier, the first formal domain model identifier, and the source event identifier. When the change event does not carry a source event identifier that meets the aforementioned conditions, the idempotent identifier is generated jointly based on the source identifier of the information source, the first target domain identifier, the first formal domain model identifier, the change operation type, the set of local identifiers or path identifiers of the affected model elements or relationships, the base model version identifier, and the normalized change content summary. The set of identifiers of the affected model elements or relationships is arranged in a preset order before participating in the generation. The operation subject, the information source identifier, a separate change object identifier, or a separate change content summary are not used individually as the idempotent identifier. When the same source event is transmitted repeatedly, the same idempotent identifier is generated, but it applies to different formal domains. Change events to the domain model, different baseline model versions, or different model elements generate different idempotency identifiers; an idempotency event index is queried to record currently being processed and already processed change events; when the same idempotency identifier already exists in the idempotency event index, the arriving change event is identified as a duplicate change event, and a duplicate event reception record is generated in an append-only immutable manner. The duplicate event reception record includes at least the original source information of the arriving event, the idempotency identifier, the reception time, the corresponding original change event identifier, and the duplication relationship; the duplicate change event does not generate new domain change evidence records, does not re-decompose atomic model changes, does not generate new semantic change candidates, and does not participate in evidence fusion. The current synchronizeability status of the corresponding formal domain model will not be revoked, restored, or changed again due to the repeated change event. The current synchronizeability status of the corresponding formal domain model will continue to be determined by the processing status of the original change event. When there is no identical idempotent identifier in the idempotent event index, the change event is registered as a non-repeating change event being processed, and the generation of domain change evidence record, temporary revocation of synchronizeability status, and subsequent processing as described in claim 1 are executed. The domain change evidence record includes at least the change source, operating entity, change time, model difference before and after the change, projection instance handle, baseline semantic version identifier, baseline model version identifier, the idempotent identifier, and a checksum for integrity verification. Verification information; when the projection instance handle is valid, the corresponding unified semantic object is directly located based on the projection instance handle and the projection relationship record; when the projection instance handle is missing, invalid, or ambiguous, the association between the semantic change candidate and the unified semantic object is re-determined according to the association matching method described in step S2; each semantic change candidate is in an evaluation state after generation, and a preliminary validity check is performed first: when the domain change evidence record on which it is based fails the integrity check, the semantic change candidate is recorded as an integrity abnormality, does not participate in evidence fusion, and does not enter the following state determination. The reason for its abnormality is recorded in the evolution field used to store the object version, change reason, and scope of influence in the dynamic semantic state field;If the baseline semantic version identifier of the domain change evidence record is inconsistent with the current semantic version, or if its baseline model version identifier is inconsistent with the version of the current formal domain model of the target domain, the semantic change candidate is marked as pending rebasing. It does not participate in evidence fusion and does not enter the following state determination. It must be re-executed based on the current semantic version, or re-calculated after rebasing or three-way comparison based on the current formal domain model version, re-decomposed atomic model changes, and regenerated semantic change candidates before the following state determination is performed. Each semantic change candidate that passes the pre-validation check is determined in the following order of priority for its subsequent state. Its candidate state set includes acceptance state, candidate reservation state, pending user confirmation state, conflict state and rejection state: (1) Semantic change candidates that violate non-exempt hard constraints after deterministic evaluation are determined to be in the rejection state; semantic change candidates that violate exemptable hard constraints after deterministic evaluation but have not yet completed a valid exemption according to the preset exemption rules are determined to be in the conflict state. After completing a valid exemption, this state determination will be re-executed. If the exemption application is rejected or the preset exemption processing conditions are exceeded, it will be converted to the rejection state; semantic change candidates that cannot be determined to violate non-exempt hard constraints due to unresolved conflicts in candidate values, scope of application of constraints or supporting evidence. (1) A semantic change candidate is identified as being in a conflict state. The semantic change candidate in the conflict state awaits supplementary evidence or constraint resolution. Before the conflict is resolved, it does not participate in evidence fusion and does not update the unified semantic object; (2) A semantic change candidate that does not violate hard constraints and meets the acceptance conditions is identified as being in an acceptance state and participates in evidence fusion. The acceptance conditions include at least: the semantic change candidate has passed the pre-validity check, its change source reliability is not lower than the preset source reliability threshold, its initial evidence support is not lower than the preset support threshold, there are no semantic ambiguities that cannot be automatically resolved, it does not violate the determined non-exempt hard constraints, and for the relevant exemptable hard constraints, its determination The result of the evaluation is that the semantic change candidate has been granted an effective exemption according to the preset exemption rules; the state of acceptance indicates that the semantic change candidate has obtained the admission to participate in evidence fusion. The fusion result is considered to be accepted semantic change and written into the determined semantic state only when the fusion result meets the convergence condition and there is no unresolved hard constraint conflict. If the convergence condition is not met after fusion, the fusion result only updates the confidence of the corresponding candidate semantic hypothesis in the hypothesis field; (3) Semantic change candidates with semantic ambiguity that cannot be automatically resolved are determined to be in the state of pending user confirmation. A confirmation request is output for presentation to the user. The user confirmation or modification result is used as a new semantic evidence record to participate in evidence fusion, but does not exempt hard constraints;(4) Semantic change candidates that have passed the preliminary validity check, do not belong to the first three items, but whose change source reliability or initial evidence support does not reach the corresponding threshold are determined to be in the candidate retention state. The candidate retention state only accepts the case where the domain change evidence record itself is valid but the evidence strength is insufficient, and does not accept the abnormal cases of integrity verification failure and the expiration of the benchmark semantic version or benchmark model version. Semantic change candidates in the candidate retention state participate in evidence fusion, but their fusion results only update the confidence of the corresponding candidate semantic hypothesis in the hypothesis field and do not write the determined semantic state. (5) Semantic change candidates in the rejection state do not participate in evidence fusion, do not update the unified semantic object, and retain the domain change evidence record, rejection state and rejection reason. After the non-repeating change event completes the status determination, corresponding disposal and the determination that the source domain model can be synchronously restored or kept in the revocation state, the status of the change event in the idempotent event index is updated from being processed to being processed, and the index relationship between its idempotent identifier and the final disposal result is retained.

8. The method according to claim 1, characterized in that: Adding, updating, or removing any projection operator, lifting operator, or corresponding conservation quantity extraction logic registered in a target domain does not change the core data structure of the unified semantic object and the dynamic semantic state field; compatibility checks, invalidation markings, or reprojections are performed on the projection relationship records and domain models affected by the addition, update, or removal, while the projection relationship records and domain models of other target projections unaffected by the addition, update, or removal remain unchanged; cross-domain synchronization and change propagation do not rely on direct point-to-point mapping between any two domain models, but are performed through the dynamic semantic state field, unified semantic object, projection operator, and lifting operator; When semantic evidence records are generated, an initial index is established between each semantic evidence fragment and its candidate semantic hypothesis. After the semantic evidence fragments are fused and accepted, a formal bidirectional index is established or updated between the semantic evidence fragment and the accepted semantic conclusion. The formal bidirectional index is used to trace back from the accepted semantic conclusion to the semantic evidence fragments that support the semantic conclusion and their respective semantic evidence records, and to locate the accepted semantic conclusion supported by the semantic evidence fragment. The reasons for domain changes, the scope of impact of changes, and the incremental reprojection results are recorded in the dynamic semantic state field to store the evolution field of object version, reasons for changes, and scope of impact.

9. The method according to claim 1, characterized in that: The target domain refers to a target model space that has its own independently registered projection operator, lifting operator, and corresponding conservation quantity extraction logic. Any two target model spaces that satisfy this condition constitute two different target domains as described in this method, regardless of whether they belong to different industries or different modeling tools. The differences between the at least two different target domains include at least one of the following: different engineering fields, different modeling language systems, different model view types under the same modeling language system, different abstraction levels, and different execution environments. The domain model includes at least one of the following: geometric model, systems engineering model, simulation model, requirements document, code skeleton, manufacturing data, and digital twin runtime model.

10. The method according to claim 1, characterized in that, Also includes: The existing domain model file is used as input data, and the existing domain model file is parsed in the initial import mode using the lifting operator registered in the corresponding target domain. The initial import mode, in the absence of records of the pre-modification model and existing projection relationships, is executed sequentially in the following two stages: In the first stage, model elements, relationships, constraints, and attribute items with object semantic content in the existing domain model file are parsed. Canvas coordinates, primitive styles, view collapse states, annotations, and editor auxiliary elements that do not carry object semantic content do not generate semantic evidence fragments. Semantic evidence fragments carrying independent positioning information, candidate semantic hypotheses, and import semantic unit source handles are generated for each of the model elements, relationships, constraints, and attribute items with object semantic content that can constitute an independent semantic processing unit (hereinafter referred to as the imported semantic unit). The connections, dependencies, and inclusion relationships between elements generate semantic evidence fragments indicating relational semantic elements. Constraint expressions and their binding relationships generate semantic evidence fragments indicating constraint semantic elements. Attributes that can be independently referred to or independently valued generate semantic evidence fragments indicating those attributes. Attributes that cannot constitute independent semantic processing units are carried along with the semantic evidence fragments of their respective model elements. Based on this, semantic evidence records are constructed and written into the dynamic semantic state field in an append-only, immutable manner. The parsing is performed according to one of the following two modes: full import mode, which parses all imported semantic units with object semantic content in the existing domain model file. In implementation, this can be achieved by importing all imported semantic units with object semantic content from the existing domain model file. The process involves comparing the existing domain model file with an empty baseline to obtain the full range of newly added differences covering the import scope. Incremental import mode only parses the model differences of the existing domain model file relative to the preset baseline model. This incremental import mode is only used when the following conditions are met simultaneously: the preset baseline model has been incorporated into the unified management of the dynamic semantic state field; each imported semantic unit shared by the preset baseline model and the existing domain model file, capable of constituting an independent semantic processing unit, already has the object identifier and corresponding projection relationship record of its associated unified semantic object; the shared imported semantic units include model elements, relationships, constraints, and elements capable of independently taking values ​​or being independently... The attribute items are referred to; for inline relations, inline constraints and inline attributes that do not constitute independent semantic processing units but are carried along with their respective model elements, their semantic content summaries and version identifiers have been incorporated into the unified management of the preset benchmark model along with their respective imported semantic units; when any common imported semantic unit lacks the object identifier, projection relation record, semantic content summary or version identifier of the associated unified semantic object, a full import mode is performed on the existing domain model file, or at least a supplementary full import is performed on the imported semantic unit category that lacks a unified management relation, so as to avoid the common parts not generating semantic evidence fragments and not being included in the unified management of the dynamic semantic state field because they do not appear in the model difference;In the second stage, after the semantic evidence fragments have completed association matching according to step S2 and have been associated with existing unified semantic objects or newly created unified semantic objects, a temporary projection relationship record between each imported semantic unit and its associated unified semantic object is established based on the source handle of the imported semantic unit carried by each semantic evidence fragment. The source handle of a model element is taken from its path or identifier in the file; the source handle of relations and constraints is taken from the element path or identifier in the file; and the source handle of an attribute item is a combination of the path or identifier of its parent model element and the identifier of the attribute field. The generated semantic evidence record is used to initialize or expand the unified semantic object so that the existing domain model is incorporated into the dynamic semantic state. Unified management of the state field; after simultaneously satisfying the following three conditions, the determination of the formal domain model of the target domain and the state transformation of the temporary projection relationship record or the separate establishment of the formal projection relationship record are completed according to the verification method adopted in the third condition: First, the semantic information items involved satisfy the convergence condition; Second, the semantic information items have independent source semantic evidence other than the existing domain model file, are confirmed by the user, and are verified by independent deterministic constraints, or the existing domain model file is explicitly marked as a trusted initialization source, wherein the marking of the trusted initialization source must simultaneously have trust basis and integrity basis: the trust basis must at least satisfy one of the following conditions, namely, confirmed by a subject with preset permissions, or verified by a trusted number. The integrity criteria include: verification of the file's signature or source credentials; confirmation through a pre-defined approval process; and belonging to a pre-registered authoritative model library. The integrity criteria must include at least content summary verification and version consistency verification, and complete source records. Files that only pass content summary verification and version consistency verification but lack the aforementioned trust criteria do not constitute a trusted initialization source. The marking subject, marking criteria, file content summary, confirmation time, and applicable scope are recorded together in the dynamic semantic state field to store the evolution field of object version, change reason, and impact scope. Existing domain model files that do not meet the above conditions are only marked as imported baseline models and are not considered independently verified semantic sources. Thirdly, the semantic conservation quantity verification passes, and this verification is performed in one of the following two ways. Execution: In the first method, the expected semantic conservation quantity or the expected model for verification of the target domain is generated based on the updated unified semantic projection range. The semantic conservation quantity comparison is performed with the existing domain model itself as the object of verification. The comparison does not overwrite the existing domain model with the generated result. When the consistency deviation is within the corresponding threshold, the existing domain model obtains a synchronizeable formal domain model state. The temporary projection relationship records corresponding to each imported semantic unit in the existing domain model file are converted into formal projection relationship records according to the convergence state, independent verification results and semantic conservation quantity verification results of their corresponding semantic information items. The imported semantic units include model elements, relationships, constraints and attribute items that can take values ​​independently or be independently referred to.For inline relations, inline constraints, and inline attributes carried along with their respective model elements, the corresponding dependency references, semantic content summaries, and version identifiers are retained in the formal projection relation record of the importing semantic unit. In the second approach, the domain model for the target domain is regenerated by re-performing formal projection from the updated unified semantic projection scope. For each generated model element, a projection relation record to be verified is first established. Then, a semantic conservation comparison is performed on the generated domain model. If the consistency deviation is within the corresponding threshold, the generated domain model is taken as the formal domain model for the target domain, and the projection relation record to be verified is converted into a formal projection relation record. Temporary projection relation records pointing to existing domain model elements are retained as import source relations and are not converted into formal projection relation records. Before all three conditions are met, the existing domain model is marked as the import baseline model and does not acquire a synchronizeable formal domain model state. If the semantic conservation verification fails, it is handled according to the verification method used: the first method is adopted... When the semantic conservation quantity verification fails, an additional failure event record is added, and the existing domain model does not acquire a synchronizeable formal domain model state. Temporary projection relationship records corresponding to each imported semantic unit in the existing domain model file are marked as failing formalization and remain in the temporary projection relationship record state. Each imported semantic unit includes model elements, relationships, constraints, and attribute items that can independently take values ​​or be independently referred to. After supplementing semantic evidence, completing independent verification, or correcting the unified semantic projection range, the verification can be re-executed. When the verification fails using the second method, an additional failure event record is added, and the generated domain model does not acquire a synchronizeable formal domain model state. The projection relationship records to be verified established for its model elements are marked as verification failed and not converted into formal projection relationship records. Temporary projection relationship records pointing to the existing domain model elements continue to be retained as imported source relationships and are not marked as invalid due to this verification failure. In both of the above situations, the corresponding historical records are not physically deleted.

11. The method according to claim 1, characterized in that, Also includes: Historical prediction candidates for future semantic states are generated based on the evolution history of the unified semantic object; the historical prediction candidates, the correction suggestion candidates generated by hard constraint violation processing, and the repair proposal candidates generated by the semantic consistency repair process are collectively referred to as restricted system candidates; The restricted system candidates are generated internally by the computer modeling system. Before obtaining independent external semantic evidence support, user confirmation, or independent deterministic verification, they do not have the identity of semantic evidence records, are not included in the number of evidence items, independent evidence related groups, or semantic evidence source categories in evidence fusion, but are used as candidate auxiliary factors independent of the semantic evidence likelihood item, and participate in the confidence calculation of the corresponding candidate semantic hypothesis with restricted weight. Each restricted system candidate must at least record the candidate identifier, candidate category, generator identifier and its version, baseline semantic version identifier, target semantic information item, generation basis, derivation chain identifier, related group identifier, idempotency identifier, generation time, and valid status. For restricted system candidates with the same baseline semantic version, the same generator version, the same target semantic information item, and semantic equivalence, replacement, merging, or discounted weighting are performed, and they are not repeatedly accumulated as mutually independent candidates. The generation or update of the restricted system candidate is not included in the number of consecutive preset number of independent evidence related group updates used in the convergence determination. When there is no unrestricted source semantic evidence for the corresponding semantic information item, the restricted system candidate participates in the confidence calculation of the candidate semantic hypothesis according to the restricted weight determined by the preset restricted benchmark weight, but shall not be used as the sole basis for the semantic information item to obtain a determined semantic state; when there is unrestricted source semantic evidence for the corresponding semantic information item, the effective weight of any restricted system candidate is lower than the lowest effective weight of the unrestricted source semantic evidence participating in the calculation of the same semantic information item. When all candidate support sources for the same semantic information item are restricted system candidates, the semantic information item shall not obtain a determined semantic state; a determined semantic state shall only be obtained after obtaining semantic evidence from unrestricted sources, or after passing independent constraint solving, simulation or deterministic verification and generating a separate semantic evidence record accordingly; no restricted system candidate shall trigger the formal generation, synchronization or incremental reprojection of the domain model on its own; after restricted system candidates are confirmed by users, independently verified or supported by external data, a separate semantic evidence record shall be generated and participate in evidence fusion in accordance with ordinary semantic evidence rules.

12. The method according to claim 1, characterized in that: The one or more information sources include external artificial intelligence models; the raw confidence information output by the external artificial intelligence models is normalized or calibrated, and the calibration results are mapped to the initial evidence support of the corresponding candidate semantic hypotheses; when the external artificial intelligence models do not provide valid confidence information, a default initial evidence support is set according to the model type, task type, and historical reliability, and this initial evidence support is recorded as the default value in the source information; the evidence source reliability weight corresponding to the external artificial intelligence models is dynamically adjusted according to their historical semantic parsing accuracy; when the output of the external artificial intelligence models participates in evidence fusion, it is subject to the constraint verification of the unified semantic object, wherein deterministic evaluation is performed. Candidate semantic hypotheses that violate non-exempt hard constraints are identified as being rejected, do not participate in evidence fusion, and do not enter a determined semantic state. Candidate semantic hypotheses that cannot be identified as violating non-exempt hard constraints due to unresolved conflicts in candidate values, scope of application of constraints, or supporting evidence are identified as being in conflict and do not participate in evidence fusion until the conflict is resolved. Candidate semantic hypotheses that violate exemptable hard constraints after deterministic evaluation but have not yet completed a valid exemption according to the preset exemption rules are also identified as being in conflict and the state determination is re-executed after a valid exemption is completed. When the exemption application is rejected or the preset exemption processing conditions are exceeded, the state is changed to rejection. Candidate semantic hypotheses that violate soft constraints suppress their confidence by the corresponding penalty coefficient. Accordingly, the candidate semantic hypotheses generated by the output of the external artificial intelligence model and the semantic change candidates generated by the domain model change are subject to the same set of hard constraint diversion rules.

13. The method according to claim 1, characterized in that: When multiple change events from one or more target domains arrive concurrently, first obtain or generate an idempotent identifier that can uniquely represent each change event; When a change event carries a source event identifier allocated by its information source in the corresponding source namespace and not reused by other change events, the idempotent identifier is generated together based on the source identifier of the information source, the corresponding target domain identifier, the corresponding formal domain model identifier, and the source event identifier. When a change event does not carry a source event identifier that meets the aforementioned conditions, the idempotent identifier is generated by combining the source identifier of the information source, the corresponding target domain identifier, the corresponding formal domain model identifier, the change operation type, the set of local identifiers or path identifiers of the affected model elements or relationships, the baseline model version identifier, and the normalized change content summary. The operation subject, the information source identifier, the individual change object identifier, or the individual change content summary are not used as the idempotent identifier on their own. The query records the idempotent event index for handling and processing change events. When the idempotent identifier of any change event is the same as an existing idempotent identifier in the idempotent event index, the change event is identified as a duplicate change event. The duplicate change event and its duplicate relationship with the original change event are recorded in an append-only immutable manner. No new domain change evidence record is generated for the duplicate change event. The current synchropotability status of its corresponding formal domain model is not revoked, restored, or changed. The duplicate change event is excluded from the current concurrent merge processing set. For each change event that is confirmed by the idempotency check to be a non-repeated change event, a domain change evidence record is generated. Each domain change evidence record carries a baseline semantic version identifier, a baseline model version identifier, an idempotency identifier, and a monotonically increasing write sequence number assigned by the dynamic semantic state field. Evidence records of changes in various fields are incorporated into evidence fusion in sequence according to the evidence time order and preset priority rules. When the evidence time is the same or incomparable, the order is determined according to the written sequence number. When the baseline semantic version identifier of the evidence record for a certain domain change is inconsistent with the current semantic version, the promotion operator processing and conflict determination are re-executed for the change event. When the baseline model version identifier of the evidence record of a change in a certain domain is inconsistent with the version of the current formal domain model of the corresponding target domain, the model difference recorded in the evidence record of the change in that domain is recalculated after performing a rebasic or three-way comparison relative to the current formal domain model version, and the atomic model change and semantic change candidate are decomposed again accordingly. For the same unified semantic projection range affected by multiple non-repeating change events, a unified determination of the scope of influence and incremental reprojection are performed after fusion. This incremental reprojection adopts a cross-target atomic transaction mode, and the model patches applied to the models of each affected domain, the updates of the corresponding projection relationship records, and the semantic conservation comparisons are treated as components of the same logical transaction. The transaction is committed as a whole only when the model patches of all affected target domains are successfully applied and the semantic conservation comparisons of each domain are passed. If the application of a model patch for any affected target domain fails or the comparison of semantic conservation quantities fails, the entire affected domain models and their projection relationship records involved in this transaction will be rolled back to the version before the patch was executed. The evidence fusion results and unified semantic object versions that have already been written will not be rolled back due to the failure of this transaction. The corresponding affected domain models will remain in an isolated state where their synchronous eligibility has been revoked, waiting for the determination of the scope of impact and incremental reprojection to be re-executed.

14. The method according to claim 1, characterized in that: The target projection set is determined by at least one of the following methods: receiving user specification; reading preset configuration; generating a recommended target projection list based on the degree of matching between the semantic information items that have met the convergence conditions in the candidate semantic range corresponding to the current modeling task and the semantic information requirements declared by each registered target projection; and adding the target projections that meet the preset recommendation conditions to the target projection set after user confirmation or preset strategy determination. The candidate semantic range is determined based on the specified root unified semantic object, the current project, user selection, or explicit specification of the modeling task. The unified semantic projection range is formed after the target projection set is determined, and newly generated recommended target projections during the execution of the current modeling task are not added to the target projection set of the current round.

15. The method according to claim 7, characterized in that, Also includes: When the association between a semantic evidence fragment and a unified semantic object is modified, the corresponding semantic evidence fragment is de-associated from the original unified semantic object and re-associated. This de-association and re-association are implemented by appending association event records, without deleting or rewriting historical association records. Each time a determined semantic state is formed through evidence fusion, the evidence set identifier supporting that determined semantic state, the semantic fusion configuration version, and the fusion result version are recorded together in the dynamic semantic state field as an evolution field for storing object versions, reasons for changes, and scope of impact. The semantic fusion configuration version is at least associated with the evidence object anchoring algorithm. The system includes rules and thresholds, evidence fusion algorithms, pre-defined prior knowledge, rules for calculating the reliability of evidence sources and the snapshot of the source reliability weights used in this instance, constraint rules, convergence thresholds and stable update times, conflict arbitration rules, evidence-related groups and discount weighting rules, and restricted system candidate processing rules. Based on the relevant semantic evidence records, semantic evidence fragments, associated event records, and recorded semantic fusion configuration versions stored in the dynamic semantic state field in an append-only immutable manner, a replay-based refusion is performed on each affected unified semantic object to reconstruct its semantic state, attributes, relationships, behavioral capabilities, constraints, and the confidence level of candidate semantic hypotheses. The replay-based re-fusion is performed according to either the historical configuration replay mode or the current configuration re-evaluation mode. In the historical configuration replay mode, the association between the corrected semantic evidence fragment and the unified semantic object is used as input. The semantic fusion configuration version corresponding to the semantic state to be replayed, the source reliability weight snapshot, and the constraint rules that were effective at the time are used to perform replay. This is used to determine the impact of association correction on the fusion result while keeping the fusion rules unchanged. For restricted system candidates, only candidate auxiliary contributions that have a valid state under the corresponding historical semantic version and whose validity can be confirmed based on the recorded candidate identifier, generator version, and baseline semantic version are restored. Restricted system candidates whose historical valid state cannot be confirmed are not included. Historical configuration replays are not included. In the current configuration re-evaluation mode, the relevant historical semantic evidence is re-fused using the currently valid semantic fusion configuration version. When re-evaluating using the current configuration, the original determined semantic state, original fusion result, and original semantic fusion configuration version are not directly overwritten. Instead, the re-evaluation result is added as a new re-evaluation event and a new unified semantic object version. The new and old semantic fusion configuration versions, result differences, reasons for re-evaluation, and scope of impact are recorded. For restricted system candidates, historical candidates that have become invalid or whose baseline semantic version does not match are not directly used. Instead, they are regenerated or re-verified according to the current generator version and then used to participate in the restricted confidence calculation of the candidate semantic hypothesis. After replay-based refusion is completed, the replay results are evaluated according to the same convergence conditions, determination thresholds, and hard constraint gating conditions as ordinary evidence fusion. When the candidate semantic hypotheses in the replay results satisfy the convergence conditions, reach the corresponding determination thresholds, and do not have unresolved hard constraint conflicts, the replay results are transformed into new accepted semantic changes, and a new unified semantic object version is generated in an append-only immutable manner. When the replay results show that the originally determined semantic content no longer satisfies the convergence conditions, no longer reaches the corresponding determination thresholds, or no longer satisfies the hard constraint gating conditions, a determined semantic withdrawal event is generated. In the new unified semantic object version, the corresponding semantic information item is changed from the determined semantic state to the candidate semantic hypothesis state or the undetermined state. The system retains the original determined semantic content, original fusion result, original semantic fusion configuration version, and its historical versions, without deleting or overwriting the historical records. When the replay result does not change the current determined semantic state, current relation state, current constraint state, or the completeness of the relevant target projection, a unified semantic state change event is not generated, and the scope of influence determination and incremental reprojection are not repeated. When the replay result generates a new accepted semantic change or determined semantic withdrawal event, or causes a change in the current relation state, current constraint state, or target projection completeness, a unified semantic state change event is generated, based on the projection relation record, relational unified semantic object, and constraint unified semantic object. The process involves determining the formal domain model that depends on the affected unified semantic object, and temporarily revoking the synchronization eligibility of the determined formal domain model; re-determining the completeness of each affected target projection; when the corresponding target projection still meets the preset model generation conditions, determining the scope of influence and performing incremental reprojection, and restoring its synchronization eligibility after the domain model obtained by incremental reprojection passes the semantic conservation quantity verification; when the corresponding target projection no longer meets the preset model generation conditions due to the determined semantic withdrawal, marking its formal domain model as semantically incomplete and maintaining the isolated state of revoked synchronization eligibility, and entering at least one of the following: active completion process, draft state domain model generation process, or regeneration process. Before the projection satisfies the preset model generation conditions again and the domain model obtained by regeneration or incremental reprojection passes the semantic conservation quantity verification, its synchronization eligibility is not restored; when a candidate for deletion of a unified semantic object obtains an accepted status and participates in evidence fusion, and its fusion result satisfies the convergence condition and there are no unresolved hard constraint conflicts that form an accepted semantic change, the unified semantic object is placed in the tombstone state; the user's confirmation or modification result of the deletion participates in evidence fusion as a new semantic evidence record and does not directly trigger the tombstone state; unified semantic objects in the tombstone state retain their object identifier, evolution history, evidence tracing reference and deletion reason, and do not participate in the current formal projection and cross-domain change propagation;For the unified semantic object, the candidate for restoration semantic change is accepted and participates in evidence fusion. After the fusion result satisfies the convergence condition and there are no unresolved hard constraint conflicts, the unified semantic object is restored to a non-tombstone state. The restoration event and its basis are recorded in the evolution field. When the unified semantic object enters the tombstone state, the relational unified semantic object and the constraint unified semantic object referencing the unified semantic object are identified based on the relation field and constraint field in the dynamic semantic state field. The tombstone change and the identified relational and constraint unified semantic objects together constitute a change set. The scope of influence of this change is determined according to the union of the relation and constraint dependencies before and after the change. The synchronization qualification of each affected formal domain model is temporarily revoked. The identified relational unified semantic objects are then set to an invalid state, and the identified constraint unified semantic objects are set to a paused evaluation state. Incremental reprojection is then performed on each model element within the affected area to generate deletion, disconnection, or deactivation patches for the unified semantic object and the corresponding model elements of the relational and constraint unified semantic objects. After the semantic conservation comparison passes, the corresponding formal domain model's synchronization eligibility is restored. Existing projection instance records of relational unified semantic objects in an invalid state and constraint unified semantic objects in a paused evaluation state are retained in the projection dependency index, used only for generating deletion, disconnection, or deactivation patches, and no longer used for subsequent ordinary semantic propagation. The invalid state indicates that the relation is temporarily not participating in the current process because its endpoint objects are unavailable. The pre-projection does not participate in the subsequent reachable closure calculation of the affected scope, but this does not mean that the relation itself has been semantically deleted; the paused evaluation state indicates that the constraint is temporarily not performing deterministic evaluation because the object it references is unavailable, and its satisfied state is not used as the basis for generation gating and propagation gating before evaluation resumes; the relational unified semantic object and the constraint unified semantic object themselves enter the tombstone state, still based on the condition that the candidate for deletion semantic change for themselves is accepted and the accepted semantic change is formed by evidence fusion, and does not directly enter the tombstone state because the endpoint object or the referenced object enters the tombstone state; when the unified semantic object is restored to the non-tombstone state, the validity of the relational unified semantic object in the invalid state and the constraint unified semantic object in the paused evaluation state is re-evaluated; Upon receiving a recovery event or recovery semantic input data for the unified semantic object, relational unified semantic object, or constraint unified semantic object, corresponding recovery semantic change candidates are generated based on the retained object evolution history, evidence tracing references, current state of endpoint objects, and constraint application conditions. The recovery semantic change candidates are subjected to state determination and evidence fusion according to the pre-validity check and state determination rules referenced in this claim. Only after an accepted semantic change is formed is the corresponding relational unified semantic object or constraint unified semantic object restored to the active state. Without the gatekeeper, it remains in an invalid state or a paused evaluation state.

16. The method according to claim 1, characterized in that: The unified semantic object exists persistently in the dynamic semantic state field, independent of any domain model; When a domain model or its elements are deleted, and this deletion, after promotion, only indicates the removal of the projection product within that target domain, the associated unified semantic object, its evidence tracing references, and evolutionary history remain valid and usable, and can be used for subsequent projection generation, evidence fusion, and cross-domain change propagation. When the semantic change candidate generated by the deletion after promotion is determined to be a semantic deletion targeting the corresponding unified semantic object, obtains acceptance status, participates in evidence fusion, and its fusion result satisfies the convergence condition and does not have any unresolved hard constraint conflicts, thus forming an accepted semantic change, the unified semantic object is placed in a tombstone state. Its object identifier, evolutionary history, evidence tracing references, and deletion reasons are retained, and it no longer participates in the current formal projection and cross-domain change propagation. When the unified semantic object enters the tombstone state, the relational unified semantic objects with it as the endpoint and the constraints referencing it remain unchanged. The same referential integrity processing is performed on relational unified semantic objects: first, the scope of influence is determined by the union of the dependencies before and after the change, and the synchronization eligibility of the affected formal domain models is temporarily revoked. Then, the relational unified semantic object is set to an invalid state, and the constraint unified semantic object is set to a paused evaluation state. Subsequently, incremental reprojection is performed on the model elements within the scope of influence to generate corresponding deletion, disconnection, or deactivation patches. After the semantic conservation comparison is passed, the synchronization eligibility of the corresponding models is restored. Relational unified semantic objects in an invalid state and constraint unified semantic objects in a paused evaluation state do not participate in the current formal projection and ordinary semantic propagation. Their object identifiers, evolution history, and evidence tracing references are preserved. Their entry into the tombstone state is still conditional on obtaining the accepted state of the deletion semantic change candidate for themselves and forming the accepted semantic change through evidence fusion.

17. The method according to claim 7, characterized in that: When the completeness of a target projection relative to the unified semantic projection range does not meet the preset model generation conditions, but meets the preset draft generation conditions corresponding to the target projection, a task version identifier is generated for the current configuration of the target projection. The task version identifier is associated with at least the semantic information items declared by the target projection, the weights and determination thresholds of each semantic information item, the formal model generation threshold and the draft generation threshold, the projection operator version registered in the corresponding target domain, the boosting operator version, the conservation quantity extraction configuration version, the normalization rule version, and the deviation threshold of each semantic conservation quantity component. A domain model marked as draft state is generated through the projection operator registered in the target domain corresponding to the target projection. When generating the domain model in the draft state, a draft projection relationship record with a draft state identifier is established. The draft projection relationship record includes at least a unified semantic object identifier or attribute item identifier, a draft model element identifier, a draft projection instance handle, and a projection operator. The system includes an identifier and version, source semantic version, draft model version, task version identifier, and mapping type. The draft projection relationship record is only used for draft model element location, draft change reverse promotion, and evidence tracing; it does not constitute a formal projection relationship record, does not grant the draft domain model synchronization eligibility, and does not participate in determining the scope of influence or incremental reprojection for the formal domain model. The draft domain model undergoes structural validity checks and non-exemptible hard constraint checks. If the checks pass, the draft domain model is presented to the user. If the checks fail, the draft domain model is not presented, but a corresponding draft generation failure message is output. The draft domain model does not participate in semantic conservation comparisons and is not marked as a synchronizationable formal domain model. For modifications to the draft domain model, a draft change evidence record is generated using the promotion operator registered in the target domain and based on the draft projection relationship record. The draft change evidence record includes at least the change source, the operating entity, the change time, the difference between the draft model before and after the change, the draft projection instance handle, the baseline semantic version identifier, the baseline draft model version identifier, the task version identifier, the idempotency identifier, and the verification information used for integrity verification; based on the draft projection relationship record, the difference is decomposed into one or more atomic model changes, and semantic change candidates are generated respectively. Each semantic change candidate undergoes state determination according to the pre-validation check, hard constraint diversion, and candidate state determination rules referenced in this claim. Specifically, the comparison of the baseline model version is based on the current draft model version and its task version identifier for the target projection, not on the version of the formal domain model of the target domain. When the baseline draft model version identifier is inconsistent with the current draft model version, the draft difference is recalculated relative to the current draft model version, the atomic model changes are re-decomposed, and semantic change candidates are regenerated before state determination is performed. When the task version changes in this round, but does not involve changes to the projection operator version, semantic information item structure, or target projection configuration, the unprocessed draft change evidence records are processed according to the new task. The version undergoes a validity check; when a task version change involves changes to the projection operator version, the structure of declared semantic information items, mapping rules, or target projection configuration, the original draft domain model and its draft projection relationship records are marked as pending reconstruction. The draft projection is then re-executed based on the new task version, and the draft projection relationship records are re-established. For draft modifications not yet submitted by the user, a three-way comparison or rebasing is performed based on the regenerated draft model before improvement, not only recalculating differences based on the old draft model. When the completeness of the target projection meets the preset model generation conditions, the formal domain model is re-generated using the updated unified semantic projection range. After the generated formal domain model passes the semantic conservation check, the corresponding... The draft state domain model and its draft projection relationship record are marked as replaced or archived, and no more change events will be received for the draft state domain model. When the user continues to edit, a new draft branch and a new draft projection relationship record will be created based on the current formal domain model or the current unified semantic projection range. The state to be reconstructed, the replaced state, and the archived state and their triggering events are recorded in the evolution field. Among them, semantic change candidates that have obtained the acceptance state participate in evidence fusion; semantic change candidates in the candidate retention state participate in evidence fusion, but their fusion results only update the confidence of the corresponding candidate semantic hypothesis in the hypothesis field; semantic change candidates in the conflict state and the rejection state do not participate in evidence fusion and do not update the determined semantic hypothesis. Semantic change candidates in the pending user confirmation state will be used as new semantic evidence records for evidence fusion after receiving user confirmation or modification results. Semantic change candidates in the candidate retention state, as well as those that have obtained the acceptance state but have not formed accepted semantic changes after fusion, will have their fusion results only update the confidence of the corresponding candidate semantic hypothesis in the hypothesis field, and the corresponding semantic information items will not be included in the completeness of the target projection. The completeness of the target projection will only be recalculated after the corresponding semantic change candidate has formed accepted semantic changes through evidence fusion and written into the determined semantic state. The domain model in the draft state does not obtain synchronization qualification, nor does it perform closed-loop verification of the semantic conservation quantity of the source domain model.When the semantic change candidate is formed into an accepted semantic change through evidence fusion and written into the determined semantic state, a unified semantic state change event is generated according to step S5 to determine the scope of influence. For each target projection affected within the scope of influence, its corresponding formal domain model's synchronizability qualification is temporarily revoked, and then incremental reprojection is performed. Only after the incremental reprojection is completed and the resulting domain model passes the semantic conservation comparison is the corresponding formal domain model's synchronizability qualification restored. The content of the draft state domain model itself is not directly applied to the domain model of any other target domain; only its accepted semantic change formed through evidence fusion is propagated to the affected target domain through the unified semantic object. When the completeness of the target projection meets the preset model generation conditions, the formal domain model is regenerated from the updated unified semantic projection range, and the content of the draft state domain model is not directly used as the content of the generated formal domain model.

18. The method according to claim 1, characterized in that: Each target projection declares a set of mandatory semantic conservation components, a set of optional semantic conservation components, a deviation threshold for each component, and a normalization rule version corresponding to its target model type. The set of mandatory semantic conservation components includes at least one or more of the following: object identity identifier, topological relationship structure, key parameters, interface signature, and effective constraint set. In step S4, the conservation quantity extraction logic performs parsing extraction with the generated domain model as input, without reusing the intermediate results generated by the projection operator during the generation of the domain model. The conservation quantity extraction logic and the projection operator can share the domain meta-model and normalization base library. The projection operator version, conservation quantity extraction configuration version, and normalization rule version are recorded respectively and stored together with the corresponding semantic conservation quantity comparison results in the dynamic semantic state field for storing the evolution field of object version, change reason, and scope of influence.

19. The method according to claim 1, characterized in that, Also includes: A task version identifier is generated for the target projection set and its configuration corresponding to this round of modeling tasks. The configuration includes the semantic information items, weights, determination thresholds, generation thresholds and draft thresholds declared by each target projection, as well as the versions of the projection operators registered in each target domain, the versions of the registered lifting operators, the versions of the conservation quantity extraction configurations, the versions of the normalization rules and the deviation thresholds of each semantic conservation quantity component. During the completeness determination, projection generation and semantic consistency verification of this round, the target projection set and its configuration remain unchanged. Newly identified candidate target projections during this round of execution are recorded in the recommended target projection list. After user confirmation or judgment by a preset strategy, they are added to the target projection set corresponding to the next task version. The control configuration inputs used to set the target projection set, various thresholds and weights, access permissions, or result presentation methods are not recorded as semantic evidence in semantic fusion. Their changes are recorded in the evolution field of the dynamic semantic state field to store the object version, the reason for the change, and the scope of influence.

20. The method according to claim 1, characterized in that: The projection relationship record can also be established using the attribute items of a unified semantic object as the mapping unit, and the mapped attribute item identifier is recorded in the projection relationship record; Mapping type is determined based on attribute items: one attribute item corresponding to one model element is recorded as one-to-one; one attribute item corresponding to multiple model elements is recorded as one-to-many; multiple attribute items jointly determine one model element and are recorded as many-to-one. The correspondence between multiple attribute items and multiple model elements is preferentially split into several independently maintainable one-to-one, one-to-many, and many-to-one records for separate storage; if reasonable splitting is not possible, it is recorded as many-to-many. When different attribute items of the same unified semantic object correspond to different model elements, their mapping type is determined separately for each attribute item, and not all are recorded as one-to-many simply because the unified semantic object as a whole corresponds to multiple model elements. The semantic evidence fragment can be associated with the unified semantic object as a whole, and can also... A specified attribute item can be associated with the unified semantic object. This attribute item has an attribute item identifier, the identifier of the unified semantic object to which it belongs, the determined semantic content of the attribute item, the candidate semantic hypothesis group and its confidence level, the current convergence state, the constraint reference and its satisfaction state, the semantic evidence fragment reference, the source semantic version, the evolution history and projection relationship. For atomic model changes of the unified semantic object, the corresponding attribute item is located based on the mapped attribute item identifier, and semantic change candidates at the attribute item granularity are generated. When an attribute item independently satisfies the convergence condition with the support of subsequent arriving semantic evidence records, its candidate semantic hypothesis has converged into an entity semantic that can be independently referred to, or it has been confirmed by evidence that it corresponds to an entity semantic with independent semantic content. When an attribute item is a component, interface, resource, or other entity of an object identity, and has at least one of the following: an independent projection relationship record, an independently recordable evolutionary history, and an independent set of constraints, the attribute item is objectified as an independent entity-type unified semantic object. Pure parameter-type attribute items that only represent values ​​and cannot be independently referred to remain as attribute items without objectification. The objectification includes: creating a new entity-type unified semantic object and assigning an independent object identifier; appending an attribute item objectification event record; and retaining the original determined semantic content, candidate semantic hypothesis groups and their confidence levels, constraint references and their satisfaction states, current convergence state, source semantic version, semantic evidence fragment references, evolutionary history, and projection relationship records associated with the attribute item. The shadow relationship record is migrated or initialized to the objectified unified semantic object, and the original attribute items are converted into references to the objectified unified semantic object without deleting its historical version. A relationship object is established between the original unified semantic object and the objectified unified semantic object. After objectification, when the objectified unified semantic object is included in the corresponding unified semantic projection range, and the corresponding target projection has declared the object identity identifier as its conserved quantity component and requires the generation of model elements corresponding to the unified semantic object, its object identity identifier is included in the object identity set of the baseline semantic conserved quantity corresponding to the target projection. The target projection that has not declared the generation of the model element does not increase its baseline object identity set due to the objectification.The migration or initialization is performed in one of the following ways: directly migrate the current determined semantic content and confidence snapshot of the attribute item and record its source semantic version, or perform a re-fusion with migration markers on the migrated semantic evidence fragments; except for this re-fusion with migration markers, the migrated semantic evidence fragments do not participate in the evidence fusion of the objectified unified semantic object again; the attribute item objectification event is also recorded as a structural change event of the unified semantic projection range in the dynamic semantic state field to store the evolution field of object version, change reason and impact range, and trigger the update of projection dependency index, the determination of impact range and the incremental reprojection of the affected target projection: before the incremental reprojection is completed and the obtained domain model is compared through semantic conservation, the synchronization qualification of each affected formal domain model is temporarily revoked; for the target projection of the existing model element that does not need to change its identifier, only update its projection relationship record so that it changes from pointing to the original attribute item to pointing to the objectified unified semantic object; for the target projection that needs to add independent model elements, generate the new model element and the associated reconnected model patch; If the target projection of the model element corresponding to the objectified unified semantic object is not declared to be unaffected by this objectification, its formal domain model's synchronizability qualification will not be revoked as a result.

21. The method according to claim 7, characterized in that, In step S5: When none of the semantic change candidates for this change have formed accepted semantic changes, the modified first formal domain model remains in a state where its synchronous qualification has been revoked and is saved as an isolated version. It is then rolled back to the original formal version before the change according to a preset strategy, awaiting supplementary evidence or manual processing before re-executing the promotion and state determination. Semantic change candidates that have not obtained an accepted state are handled according to their respective states. Semantic change candidates that have obtained an accepted state but have not met the convergence conditions after fusion have their candidate semantic hypotheses retained in the hypothesis field for subsequent evidence to converge. Their corresponding model elements, relationships, or attribute items are subject to partial rollback or isolation marking. When some semantic change candidates form accepted semantic changes, while the remaining semantic change candidates are in a rejection state, conflict state, candidate retention state, pending user confirmation state, or although they have obtained an accepted state, they have not yet formed accepted semantic changes after fusion, the unified semantic object is updated only based on the accepted semantic changes, and handled according to their respective states: For semantic change candidates in a rejection state, the corresponding model elements, relationships, or attribute items are subject to partial rollback or isolation marking; for semantic change candidates in a conflict state... The corresponding model elements, relationships, or attribute items are isolated and await supplementary evidence or constraint resolution; user confirmation requests are output for semantic change candidates in the pending user confirmation state; for semantic change candidates in the candidate retention state or those that have been accepted but have not yet formed accepted semantic changes after fusion, their candidate semantic hypotheses are retained in the hypothesis field, and their corresponding model elements, relationships, or attribute items are partially rolled back or isolated; before all semantic change candidates related to this change have been processed, the modified first formal domain model will not regain its synchronization eligibility; the processing is completed when all semantic change candidates related to this change have reached one of the following final conditions: first, accepted semantic changes are formed through evidence fusion and written into the determined semantic state; second, they are determined to be in the rejection state, and their corresponding model elements, relationships, or attribute items have been partially rolled back or isolated; third, they are determined to be in the candidate retention state or have been accepted but have not formed accepted semantic changes after fusion, their candidate semantic hypotheses are retained in the hypothesis field, and their corresponding model elements, relationships, or attribute items have been partially rolled back or isolated; For semantic change candidates that are in the pending user confirmation state, the status determination must be re-executed after receiving user confirmation or modification result, and the process is considered complete only when the result of the re-determination meets one of the aforementioned three conditions. For semantic change candidates in a conflict state, the state determination must be re-executed after the conflict is resolved, and the process is considered complete only when the result of the re-determination meets one of the aforementioned three conditions. Semantic change candidates that are marked as having integrity abnormalities or pending rebasing and have not entered the state determination after the aforementioned pre-validation check must undergo state determination again after rebasing and re-promotion and meet one of the aforementioned three conditions, or their corresponding model elements, relationships, or attribute items have undergone partial rollback or isolation marking, before they are considered to have been processed. Semantic change candidates that are still in a pending user confirmation state or a conflict state after re-execution of the state determination are not considered to have been processed, and the corresponding first formal domain model after the change will not regain its synchronization eligibility.

22. The method according to claim 21, characterized in that, In step S5: when there is an accepted semantic change, regardless of whether there is an affected second target domain projection in the scope of influence, a verification expectation model or expected semantic conservation quantity of the first target domain is generated according to the updated unified semantic projection scope. The current valid semantic view of the first formal domain model after excluding model elements, relationships and attribute items that have been isolated is used as the verification object to perform semantic conservation quantity closed-loop verification. The closed-loop verification does not cover the changed first formal domain model with the verification expectation model. The changed first formal domain model is restored to synchronization eligibility only when all semantic change candidates of this change have been processed as described in claim 21 and the consistency deviation of the closed-loop verification is within the corresponding threshold. When the consistency deviation exceeds the corresponding threshold, a repair proposal is generated or a manual confirmation request is output. The result of the manual confirmation is used as a semantic evidence record, a repair proposal selection record, or a control configuration input to enter the corresponding processing flow. Synchronization eligibility is restored only after evidence fusion, projection, or domain model repair and semantic conservation quantity closed-loop verification are successfully performed. The manual confirmation itself does not directly restore synchronization eligibility for domain models whose consistency deviation exceeds the corresponding threshold. After determining one or more affected second target domain projections, the synchronization eligibility of the corresponding formal domain models in each second target domain is temporarily revoked, and the synchronization eligibility of each model is restored only after the corresponding incremental model patch is successfully applied and its semantic conservation quantity comparison passes. If the application of any second target domain incremental model patch fails or the semantic conservation quantity comparison fails, the second target domain... The domain model content and projection relationship records of the target domain are rolled back to the version before the patch was executed, and are kept in an isolated state or a consistency check failure state without restoring their synchronization eligibility before this incremental reprojection, until the incremental reprojection is re-executed for the second target domain and passes the semantic conservation comparison. Other second target domains that have passed the semantic conservation comparison will not have their restored synchronization eligibility revoked. The model updates actually applied to each second target domain model by the incremental reprojection are respectively attached with system source update identifiers. No new domain change evidence records are generated for model update events carrying valid system source update identifiers. The verification expected model or expected semantic conservation quantity is only used as a verification product within the computer modeling system. It is not written into the first formal domain model, does not enter the domain model change event listening channel, and is not used as a system source update.

23. A cross-domain modeling system for unified semantic objects, characterized in that, The system includes at least one processor and a memory, the memory storing computer instructions; the system has projection operators and lifting operators registered for at least two target domains, and conservation quantity extraction logic configured for each target domain; when the at least one processor executes the computer instructions, it implements the following functional modules: an evidence generation module, used to acquire input data from one or more information sources, generate semantic evidence records based on semantic input data used to determine the semantic content of the object in the input data, the semantic evidence records including at least source information and one or more semantic evidence fragments, each semantic evidence fragment including location information, candidate semantic hypotheses and initial evidence support corresponding to each candidate semantic hypothesis, and the evidence generation module... Semantic evidence records are stored in the dynamic semantic state field in an append-only, immutable manner. Semantic input data from any information source participates in the updating of the unified semantic object in the form of semantic evidence records, without directly overwriting the already determined semantic state of the unified semantic object. The determined semantic state includes at least one of the following: semantic state, attributes, relationships, behavioral capabilities, and constraints that have met convergence conditions. A completeness determination module is used to obtain the target projection set corresponding to this round of modeling tasks. The target projection set contains one or more target projections corresponding to each registered target domain. A semantic object association and fusion module is used to associate and match each semantic evidence fragment with an existing unified semantic object in the dynamic semantic state field, on a unit basis. Semantic objects are categorized into entity-type unified semantic objects, relational-type unified semantic objects, and constraint-type unified semantic objects based on the types of semantic elements they carry. Relational-type unified semantic objects carry type-based relationships between two or more unified semantic objects, while constraint-type unified semantic objects carry verifiable constraint expressions acting on one or more unified semantic objects and their satisfied states. The semantic object association and fusion module is further used to associate semantic evidence fragments with any existing unified semantic object of any type based on the association matching results, or to create new entity-type, relational-type, or constraint-type unified semantic objects and establish associations based on the semantic element type indicated by the semantic evidence fragment. Different semantic elements in the same semantic evidence record... Semantic evidence fragments can be associated with different unified semantic objects; one or more semantic evidence fragments associated with the same unified semantic object are fused, and at least one of the semantic state, attributes, relationships, behavioral capabilities, and constraints of the unified semantic object is updated, and semantic content that has not yet met the convergence condition is retained as a candidate semantic hypothesis state without being written into the determined semantic state; the semantic object association and fusion module is also used to form a corresponding unified semantic projection range according to the target projection set, the unified semantic projection range including one or more entity-type unified semantic objects, and at least one of relation-type unified semantic objects associated with the entity-type unified semantic objects and constraint-type unified semantic objects acting on the entity-type unified semantic objects;When no relational or constraint-type unified semantic objects exist, the unified semantic projection range is composed of one or more entity-type unified semantic objects. The completeness determination module is further configured to determine the completeness of the unified semantic projection range relative to each target projection based on the semantic information requirements declared by each target projection. For target projections whose completeness does not meet the preset model generation conditions, the formal domain model generation for that target projection in this round is stopped, or at least one of the active completion process or the draft state domain model generation process is entered. The projection and consistency verification module is used to, for target projections whose completeness meets the preset model generation conditions, use the projection operator registered in the target domain corresponding to that target projection to determine the completeness of the unified semantic projection range. The projection generates a corresponding domain model, and establishes and stores projection relationship records between semantic elements in the unified semantic projection range and model elements in the generated domain model. The semantic elements include entity-type unified semantic objects, relation-type unified semantic objects, and constraint-type unified semantic objects in the unified semantic projection range. The domain models corresponding to at least two different target domains are generated through one or more projections, and are included in cross-domain change propagation after the domain models of at least two different target domains have obtained synchronization eligibility. Each target projection declares the set of mandatory semantic conservation components, the set of optional semantic conservation components, the deviation threshold corresponding to each semantic conservation component, and the normalized rule version corresponding to its target model type. The required set of semantic conservation components includes at least object identity identifiers, and at least one of topological relationship structure, key parameters, interface signatures, and effective constraint sets. The projection and consistency verification module is further configured to obtain baseline semantic conservation quantities corresponding to the semantic conservation components declared by each target projection from the unified semantic projection range, and through conservation quantity extraction logic set corresponding to each target domain, using at least one of the persistent model content, executable structure, or parsable model structure of the generated domain model as input, and along a model parsing path independent of the corresponding projection operator, to parse and obtain normalized semantic conservation quantities from each domain model respectively. The conservation quantity extraction logic does not reuse the projection operator in the process of generating the corresponding domain model. The semantic conservation values ​​generated directly or the intermediate results used to construct the semantic conservation values ​​are used. The conservation extraction logic and the projection operator can share the domain meta-model and normalization base library of the corresponding target domain. The projection and consistency verification module is also used to compare the mandatory semantic conservation components and the actual optional semantic conservation components obtained from parsing each domain model with the corresponding components in the benchmark semantic conservation. The absence of components that are not declared as mandatory semantic conservation components by the corresponding target projection does not constitute a semantic consistency verification failure. Based on the consistency deviation obtained from the comparison, it is determined whether the corresponding domain model meets the semantic consistency condition, and only the domain model that meets the semantic consistency condition is marked as a synchronizable formal domain model.The projection and consistency verification module is also used to attach a system source update identifier to the model update actually applied to the domain model by incremental reprojection, and to associate and store the system source update identifier with the corresponding unified semantic state version identifier, the updated target model version identifier, and the update operation record in the evolution field of the dynamic semantic state field for storing object version, change reason, and scope of influence; only when the system source update identifier and its associated information carried by the model update event are consistent with the corresponding record in the evolution field is the system source update identifier determined to be a valid system source update identifier; the verification expectation model or used for the closed-loop verification of the semantic conservation quantity of the source domain model is used. The system expects semantic conservation quantities to be used only as internal verification products, not written into the corresponding formal domain model, not entered into the domain model change event listening channel, and not used as system source updates. The change evidence and promotion module, upon receiving a change event of a synchronizeable first formal domain model in the first target domain that does not carry a valid system source update identifier, generates an append-only immutable domain change evidence record based on the change event, and temporarily revokes the synchronizeability of the changed first formal domain model, preventing it from participating in cross-domain change propagation until the change processing is complete. Through the promotion operator registered in the first target domain, the system further promotes the domain change evidence based on the domain change... The evidence record, the model difference before and after the change of the first formal domain model, and the projection relationship record are used to decompose the model difference into one or more atomic model changes, and generate semantic change candidates involving one or more semantic elements in the unified semantic projection range respectively; the change evidence and boosting module is also used to, when receiving a model update event carrying a valid system source update identifier, not generate a new domain change evidence record for the model update event, and not perform boosting operator processing and evidence fusion; when the system source update identifier carried by the model update event cannot be consistent with the corresponding record in the evolution field, it is treated as if it does not carry a valid system source update identifier. Change event processing; the change evidence and enhancement module is also used to perform state determination and corresponding processing on each semantic change candidate. Among them, semantic change candidates that have obtained the acceptance status through state determination are associated with the domain change evidence record as candidate semantic interpretations and participate in evidence fusion. Only when the fusion result meets the convergence condition and there is no unresolved hard constraint conflict that forms an accepted semantic change, the fusion result is used to update the determined semantic state of the unified semantic object. Semantic change candidates that have not obtained the acceptance status, as well as semantic change candidates that have obtained the acceptance status but have not formed an accepted semantic change after fusion, do not update the determined semantic state.The projection and consistency verification module is further used to, after the change evidence and enhancement module determines the semantic change candidates and forms accepted semantic changes through evidence fusion, and before performing the impact scope determination and source domain model semantic conservation closed-loop verification, perform the following: based on the atomic model change type corresponding to the semantic change candidates, the unified semantic object positioning result obtained directly based on the valid projection instance handle and projection relationship record, or obtained through association matching when the projection instance handle is missing, invalid, or ambiguous, and the updated unified semantic projection range, perform formal projection verification on the model elements, association relationships, or attribute items in the first formal domain model and the corresponding unified semantic object or the corresponding unified semantic object attribute item. The shadow relationship record is marked for establishment, update, termination, or invalidation. Specifically, for new model elements, new associations, or new constraints, a projection instance handle is generated for the corresponding model element, association, or constraint, and a formal projection relationship record is established containing the corresponding unified semantic object identifier or attribute item identifier, source model local identifier, mapping type, source semantic version, current version of the first formal domain model, projection operator identifier and version, and promotion operator identifier and version. For updates to existing model elements or attribute items, at least one of the source semantic version, target model version, and mapped attribute item identifier in the corresponding formal projection relationship record is updated. For association reconnection, the corresponding relational unified semantic object is updated. The endpoint references of semantic objects and their formal projection relationship records; for the deletion of model elements, relationships, or constraints, depending on whether the accepted semantic change represents the semantic deletion of a unified semantic object or only the removal of projection artifacts in the source target domain, the corresponding formal projection relationship record is marked as terminated or invalid; the establishment, updating, termination, or invalidation of the formal projection relationship record is all implemented by adding projection relationship change events and retaining historical versions; only after the relevant formal projection relationship record has completed the aforementioned processing, the corresponding model element, relationship, or attribute item can enter the current valid semantic view of the first formal domain model as the currently valid projection instance; when the formal projection relationship record processing fails, it is not... The system rolls back accepted semantic changes that have already been formed, and marks the corresponding projection relationship record as pending reconstruction, so that the first formal domain model remains eligible for synchronization until the projection relationship record is reconstructed and passes the semantic conservation quantity closed-loop verification. Semantic change candidates that have not formed accepted semantic changes do not establish or activate new formal projection relationship records. The change evidence and enhancement module is also used to perform at least one of the following on the model elements, relationships or attribute items in the first formal domain model corresponding to the semantic change candidate that have not formed accepted semantic changes, based on their status: partial rollback, isolation marking, waiting for user confirmation, waiting for supplementary evidence, conflict resolution or rebasing.The isolation marker is used to prevent the corresponding model elements, relationships, or attribute items that carry the semantic content of the object from entering the current valid semantic view of the first formal domain model. Model elements, relationships, and attribute items in the isolation marker state are retained in the isolation version and do not participate as currently valid projection instances in semantic conservation extraction, cross-domain change propagation, and model difference benchmarks of subsequent formal domain model versions. The current valid semantic view consists of model elements, relationships, attribute items, and constraints that are not in the isolation state, exist as currently valid projection instances, and carry the semantic content of the object. Canvas coordinates, primitive styles, view collapse states, annotations, and editing elements that do not carry the semantic content of the object are excluded. Editor auxiliary elements do not belong to the model semantic content that needs to be supported by the currently determined semantic state; among them, semantic change candidates determined to be in a rejected state are considered to have completed processing only after their corresponding model elements, relationships, or attribute items have undergone partial rollback or isolation marking; semantic change candidates determined to be in a candidate retention state, and semantic change candidates that have obtained the acceptance state but have not yet formed an accepted semantic change after fusion, are considered to have completed processing only after their candidate semantic assumptions are retained in the hypothesis field of the dynamic semantic state field, and their corresponding model elements, relationships, or attribute items have undergone partial rollback or isolation marking; semantic change candidates in a pending user confirmation state or conflict state After receiving user confirmation or modification results, supplementary evidence, or conflict resolution, the status determination is re-executed. The process is considered complete only if, after re-determination, an accepted semantic change is identified, or a rejection status is determined and a partial rollback or isolation marker is completed for the corresponding model element, relationship, or attribute item, or a candidate retention status is determined and a partial rollback or isolation marker is completed for the corresponding model element, relationship, or attribute item. Semantic change candidates that did not enter the status determination process due to integrity verification anomalies or inconsistencies between the baseline semantic version and the baseline model version will only be considered complete if the status determination is re-executed after correction, rebasing, or re-promotion, and any of the aforementioned completion determinations are achieved. Alternatively, the processing is considered complete only after the corresponding model element, relationship, or attribute item has undergone partial rollback or isolation marking; the impact analysis and incremental reprojection module is used only when the semantic change candidate is formed by evidence fusion to meet the convergence condition and there are no unresolved hard constraint conflicts accepted semantic changes, to determine the impact range of the change event based on the relationships and constraints in the updated unified semantic projection range, and to determine the range of affected model elements based on the projection relationship record; when there are one or more affected second target domain projections in the impact range, an incremental reprojection request for each affected second target domain projection is sent to the projection and consistency verification module;The projection and consistency verification module is further configured to, upon receiving the incremental reprojection request, first temporarily revoke the synchronization eligibility of the corresponding formal domain models of each affected second target domain, and then, based on the updated unified semantic projection range and the projection operators registered by each second target domain, perform incremental reprojection on the projections of each affected second target domain. Synchronization eligibility of the corresponding formal domain model is restored only after the corresponding incremental model patch is successfully applied and the resulting domain model passes the semantic conservation comparison. If the incremental model patch application of any second target domain fails or the semantic conservation comparison fails, the domain model content and its projection relationship records of that second target domain are rolled back to the version before the execution of the incremental model patch. This process maintains the domain model of the second target domain in an isolated state where its synchronization eligibility has been revoked or a consistency check has failed, until incremental reprojection is re-executed for the second target domain and passes the semantic conservation comparison. Formal domain models in other second target domains that have passed the semantic conservation comparison are not revoked of their restored synchronization eligibility. The model update actually applied to the domain model by the incremental reprojection is appended with the system source update identifier. The impact analysis and incremental reprojection module is also used to fuse semantic evidence records generated from semantic input data from any information source to form accepted semantic changes that cause changes in the determined semantic content within the unified semantic projection range, and these changes are not caused by synchronized... When a change event occurs in the formal domain model of a step, a unified semantic state change event is generated; the scope of influence of the unified semantic state change event is determined based on the relationships and constraints in the updated unified semantic projection range and the projection relationship record, and an incremental reprojection request is sent to the projection and consistency verification module for each affected target projection within the scope of influence; the projection and consistency verification module is further configured to temporarily revoke the synchronization eligibility of the formal domain model corresponding to each affected target projection before performing the incremental reprojection triggered by the unified semantic state change event, and restore its synchronization eligibility only after the corresponding incremental reprojection is completed and the resulting domain model passes the semantic conservation comparison; in the unified semantic state... When a state change event is not caused by a synchronous formal domain model change event, the semantic conservation quantity closed-loop verification of the source domain model is not performed; the projection and consistency verification module is also used to generate the verification expectation model or expectation semantic conservation quantity of the first target domain based on the current unified semantic projection range after all semantic change candidates related to the current change of the first formal domain model have completed the aforementioned processing, regardless of whether an accepted semantic change has been formed. When an accepted semantic change is formed, the current unified semantic projection range is the unified semantic projection range updated based on the accepted semantic change. When an accepted semantic change is not formed, the current unified semantic projection range retains the determined semantic content before the occurrence of the current change event.Using the current valid semantic view of the first formal domain model as the object to be verified, a semantic conservation loop verification of the source domain model is performed. This semantic conservation loop verification does not cover the modified first formal domain model with the expected verification model. The change evidence and enhancement module is further configured to only verify model elements, relationships, or attribute items that are related to the current change and carry object semantic content, whose semantic content is not supported by the currently determined semantic state and corresponding formal projection relationship records, and for which local rollback or isolation marking has not been performed, and whose semantic conservation... When the consistency deviation obtained from the constant quantity closed-loop verification is within the corresponding threshold, the synchronizeability of the modified first formal domain model is restored; when the consistency deviation exceeds the corresponding threshold, a repair proposal is generated or a manual confirmation request is output, and the synchronizeability of the first formal domain model is restored only after the evidence fusion, projection, or domain model repair and the semantic conservation quantity closed-loop verification are successfully performed again; the semantic object association and fusion module is also used to record the evidence set identifier supporting the determined semantic state, the semantic fusion configuration version, and the fusion result version in the evolution field each time a determined semantic state is formed by evidence fusion; the semantic fusion configuration version is at least associated with semantic evidence pieces. The algorithm for associating and matching segments with unified semantic objects, including thresholds, evidence fusion algorithms, pre-defined prior knowledge, rules for calculating the reliability of evidence sources and the snapshot of source reliability weights used in this instance, constraint rules, convergence thresholds and stable update counts, conflict arbitration rules, evidence-related groups and discount weighting rules, and rules for processing candidate restricted systems. The semantic object association and fusion module is also used to, when the association between semantic evidence segments and unified semantic objects is modified, remove the corresponding semantic evidence segment from the original associated unified semantic object and re-establish the association. The removal and re-establishment of association are implemented by appending association event records, without deleting or rewriting historical association records. Based on the dynamic... The semantic state field stores relevant semantic evidence records, semantic evidence fragments, associated event records, and recorded semantic fusion configuration versions in an append-only immutable manner. Replay-based refusion is then performed on each affected unified semantic object to reconstruct its semantic state, attributes, relationships, behavioral capabilities, constraints, and the confidence level of candidate semantic hypotheses. The replay-based refusion is performed according to either a historical configuration replay mode or a current configuration re-evaluation mode. In the historical configuration replay mode, the association between the corrected semantic evidence fragments and the unified semantic objects is used as input, and replay is performed using the semantic fusion configuration version corresponding to the semantic state to be replayed, the source reliability weight snapshot, and the then-currently valid constraint rules.In the current re-evaluation mode, the relevant historical semantic evidence is re-fused using the currently valid semantic fusion configuration version. This does not directly overwrite the original determined semantic state, the original fusion result, or the original semantic fusion configuration version. Instead, the re-evaluation result is appended as a new re-evaluation event and a new unified semantic object version, recording the old and new semantic fusion configuration versions, result differences, reasons for re-evaluation, and scope of impact. The semantic object association and fusion module is also used to determine the replay-based re-fusion result according to the same convergence conditions, determination thresholds, and hard constraint gating conditions as ordinary evidence fusion. When the candidate semantic hypothesis in the replay-based re-fusion result satisfies the convergence condition and reaches the corresponding determination threshold, and... When there are no unresolved hard constraint conflicts, the replay-based refusion result is used to form a new accepted semantic change, and a new unified semantic object version is generated in an append-based immutable manner. When the replay-based refusion result shows that the originally determined semantic content no longer meets the convergence condition, no longer reaches the corresponding determined threshold, or no longer meets the hard constraint gating condition, a determined semantic withdrawal event is generated. In the new unified semantic object version, the corresponding semantic information item is changed from the determined semantic state to the candidate semantic hypothesis state, the undetermined state, or the conflict state, and the original determined semantic content, the original fusion result, the original semantic fusion configuration version, and its historical versions are retained. The impact analysis and incremental reprojection module is also used to perform replay-based refusion... When a new accepted semantic change or confirmed semantic withdrawal event is formed through fusion, or when the current relation state, current constraint state, or target projection completeness changes, a unified semantic state change event is generated. Based on the updated unified semantic projection range, projection relation records, relational unified semantic objects, and constraint unified semantic objects, the affected target projections are determined, and the completeness of each affected target projection is re-determined. For target projections that still meet the preset model generation conditions after re-determination, an incremental reprojection request is sent to the projection and consistency verification module. For target projections that no longer meet the preset model generation conditions due to confirmed semantic withdrawal, the corresponding formal domain model is withdrawn from synchronization. The domain model is marked as having an incomplete semantic baseline until the target projection meets the preset model generation conditions again, and the domain model obtained by regeneration or incremental reprojection passes the semantic conservation comparison before it regains its synchronization eligibility. When the replay-based refusion result does not change the currently determined semantic state, current relation state, current constraint state, and the completeness of the relevant target projection, a unified semantic state change event is not generated, and the scope of influence determination and incremental reprojection are not repeated. The system is also used to isolate and save domain models that have not obtained synchronization eligibility, whose synchronization eligibility has been revoked, are in a state of incomplete semantic baseline, or whose semantic consistency verification has failed, so that they do not participate in the automatic cross-domain synchronization link.In this context, the first target domain differs from the second target domain. Cross-domain change propagation is performed via the unified semantic projection range and projection and lifting operators in the dynamic semantic state field, rather than directly applying changes from the first formal domain model to the domain model of the second target domain.

24. The system according to claim 23, characterized in that: The projection and consistency verification module is also used to establish projection relationship records with the attribute items of the unified semantic object as the mapping unit. The mapping relationship records record the identifier of the mapped attribute item and determine the mapping type based on the attribute item: when one attribute item corresponds to one model element, it is recorded as one-to-one; when one attribute item corresponds to multiple model elements, it is recorded as one-to-many; when multiple attribute items jointly determine one model element, it is recorded as many-to-one. The correspondence between multiple attribute items and multiple model elements is preferentially split into several independently maintainable one-to-one, one-to-many, and many-to-one records and stored separately. When it cannot be reasonably split, it is recorded as many-to-many. When different attribute items of the same unified semantic object correspond to different model elements, they are determined separately according to each attribute item. Based on this, a projection dependency index that can be retrieved with the attribute item identifier as the key is constructed. The change evidence and enhancement module is further used to locate the atomic model change to the corresponding attribute item based on the mapped attribute item identifier, and generate semantic change candidates at the attribute item granularity; the semantic object association and fusion module is further used to create a new entity-type unified semantic object when an attribute item independently satisfies the convergence condition, its candidate semantic hypothesis has converged to an entity semantic that can be independently referred to, or is confirmed by evidence to correspond to a component, interface, resource or other entity with an independent object identity, and has at least one of the following: an independent projection relationship record, an independently recordable evolutionary history and an independent constraint set. Pure parameter-type attribute items that only represent values ​​and cannot be independently referred to are retained as attribute items and not created as entity-type unified semantic objects; the module migrates or initializes the determined semantic content, candidate semantic hypothesis group and its confidence, constraint reference and its satisfaction status, current convergence status, source semantic version, semantic evidence fragment reference, evolutionary history and projection relationship record corresponding to the attribute item to the created one. The unified semantic object is migrated or initialized in one of two ways: either by directly migrating the current determined semantic content and confidence snapshot of the attribute item and recording its source semantic version, or by performing a refusion with migration markers on the migrated semantic evidence fragments. Except for the refusion with migration markers, the migrated semantic evidence fragments are not repeatedly included in the evidence fusion of the created unified semantic object. The original attribute item is converted into a reference to the unified semantic object without deleting its historical version, and a relationship object is established between the original unified semantic object and the created unified semantic object. The projection and consistency verification module is also used to update the projection dependency index after the attribute item is objectified, and temporarily revoke the synchronization qualification of each affected formal domain model before the incremental reprojection of the affected target projection is completed and the semantic conservation quantity comparison passes. For existing model element identifiers that do not need to be changed, only the object pointed to by its projection relationship record is updated. For those that need to add independent model elements, a new model element and a model patch for reconnection are generated.

25. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 22.

26. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to implement the method of any one of claims 1 to 22.

Images