Complex system comprehensive multi-view consistency detection method based on large model

Abstract

The invention discloses a complex system comprehensive multi-view consistency detection method based on a large model, and relates to the technical field of information system architecture design. The method comprises the steps that N view models to be detected are determined, and a rule base is constructed and formed; forming a triple set; performing rule retrieval on the triple set to obtain a corresponding rule subset; and constructing a complete prompt statement according to the prompt template and the rule subset, and reasoning the prompt statement by the large language model to obtain a consistency detection result. Based on the MBSE method, the ability of understanding, reasoning and applying knowledge in the field of system architecture and system engineering is remarkably improved, the automation degree, accuracy and robustness of multi-view consistency detection are improved, and the correctness, completeness and consistency of complex information system architecture design are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of information system architecture design technology, specifically to a method for synthesizing multiple perspectives in complex systems based on a large model. Figure 1 Infectivity testing methods. Background Technology

[0002] Model-Based Systems Engineering (MBSE) is a systems engineering methodology that emphasizes using models to drive the entire lifecycle of a system, including requirements definition, design, verification, and validation. SysML (Systems Modeling Language) is a system modeling language that supports MBSE implementation. It provides nine types of views, including sequence diagrams, state machine diagrams, parameter diagrams, package diagrams, and requirement diagrams. It offers a complete set of systematic and structural methods and a graphical representation. SysML includes module definition diagrams, internal module diagrams, use case diagrams, activity diagrams, and temporal modeling tools, enabling it to comprehensively express multi-perspective information such as the system's structure, behavior, parameters, and constraints.

[0003] In systems engineering, various views and their corresponding data models characterize system content from different dimensions, effectively enhancing the completeness and richness of the architecture expression. However, this also brings problems such as complex relationships between models and difficulties in management. Especially in practice, inconsistencies often arise between different views due to data overlap and mutual references. While architecture design tools like DoDAF (such as source diagrams) and system design tools like MBSE's SysML design tools (such as SysDeSim.Arch) can detect errors in view standard format specifications and basic syntax under general rules during the engineering design process, they have certain limitations and can only identify general, simple inconsistencies; they are far from meeting the needs of multi-view design in the current complex information system design process. Figure 1 Consistency verification requirements. Traditional architecture design verification methods struggle to accurately identify logical relationships across components and projects, making it difficult to meet the cross-view verification needs of complex information system architectures. Figure 1 Infection detection often has a large blind spot, resulting in unsatisfactory detection results. Summary of the Invention

[0004] This invention provides a multi-view synthesis method for complex systems based on a large model. Figure 1 In order to solve the problems of multi-view detection methods in existing technologies Figure 1 The problem of low accuracy in contamination testing.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A Multi-View Synthesis for Complex Systems Based on a Large Model Figure 1 Inconsistency testing methods include:

[0007] N view models to be detected are identified. A structured initial view specification is constructed based on the identified view models. A large language model is used to perform semantic induction and knowledge reorganization on the initial view specification to construct a rule base.

[0008] The data from N view models are normalized and exported according to the structured XML file. The normalized XML file is parsed using regular expression matching to form a set of triples.

[0009] The rule set of triples is retrieved by the rule base to obtain the corresponding rule subset;

[0010] Obtain a prompt template, construct a complete prompt statement based on the prompt template and the rule subset, input the prompt statement into the large language model, and the large language model performs reasoning on the prompt statement to obtain the consistency detection results of N view models;

[0011] Where N > 1.

[0012] Based on this, the present invention can be further improved as follows:

[0013] The initial view specification is semantically summarized and knowledge reorganized using a large language model to construct a rule base, specifically including:

[0014] The large language model parses the initial view specification, analyzes the semantic association, structural mapping and usage patterns of the modeling elements among the N view models, and extracts a set of consistency constraint rules with universality and interpretability.

[0015] Each consistency constraint rule is centered around a relation-view pair and is represented as relation-view-rule. The relation describes the semantic or structural association between modeling elements, the view describes the view type of the view model to which the consistency constraint rule applies, and the rule describes the consistency constraints that the relation and the view must satisfy.

[0016] The generated consistency constraint rules are represented in vector form using a text encoder, and all vectors are stored in the database to form a rule base.

[0017] Based on this, the present invention can be further improved as follows:

[0018] After normalizing and exporting the data from N view models according to the structured XML file format, it also includes:

[0019] Unnecessary namespace nodes are removed from the exported XML file, and then the XML processing tool is used for formatting, alignment, and line breaks.

[0020] Based on this, the present invention can be further improved as follows:

[0021] The normalized exported XML file is parsed using regular expression matching to form a set of triples, specifically including:

[0022] Construct a tree structure corresponding to the XML file, extract model element information, including: nodes, packages, classes, relationships and their attributes, and identify the feature IDs and related attribute information of each node and relationship based on the model layer data;

[0023] Parse the graphical structure information associated with the view to obtain a structural description of the connection relationships between nodes, wherein the connection relationship is in the form of first entity ID - relationship ID - second entity ID;

[0024] By combining model layer data and graphical structure information, the entity relationships existing in each view model are systematically reconstructed, and finally a formalized set of triples is generated.

[0025] Based on this, the present invention can be further improved as follows:

[0026] The system refactors the entity relationships existing in each view model, specifically including:

[0027] Based on the ID representation in the graphical information, find the relevant element ID in the model layer data, and replace the ID representation in the original connection relationship representation with the specific names of the nodes and relationships.

[0028] Based on this, the present invention can be further improved as follows:

[0029] The rule base is used to perform rule retrieval on the set of triples to obtain the corresponding rule subset, specifically including:

[0030] By using the view name and relation name corresponding to the triple set, a text encoder is used to obtain the corresponding embedded code. The embedded code is then used to retrieve the rule base to obtain the corresponding preliminary rule subset.

[0031] The rule with the highest similarity in the initial rule subset is determined and returned as the rule for the current set of triples;

[0032] The set returned by all triples is a subset of the rules that the view needs to follow.

[0033] Based on this, the present invention can be further improved as follows:

[0034] The rule base is retrieved using the following formula:

[0035]

[0036] Where Similarity represents similarity, A represents the embedded encoding of the triple set, and B represents the embedded representation of the rules in the rule base.

[0037] Based on this, the present invention can be further improved as follows:

[0038] The prompt templates include:

[0039] Task instructions: Describe the consistency detection task that the large language model needs to complete;

[0040] Contextual information: Describes the background content, including input and output examples;

[0041] Input data: The extracted set of triples and the corresponding rule subsets are embedded into the prompt content;

[0042] Output requirements: Guide the large language model to generate response results that conform to the expected format.

[0043] Based on this, the present invention can be further improved as follows:

[0044] The large language model infers from the prompt statement to obtain consistency detection results for N view models, specifically including:

[0045] The large language model infers the prompt statement, and when the data analysis of the N view models of the prompt statement is consistent, the output consistency detection result is consistent.

[0046] When the data analysis of the N view models in the prompt statement is inconsistent, the consistency check result output is inconsistent, and the output result includes an explanation and description of the inconsistent content.

[0047] Based on this, the present invention can be further improved as follows:

[0048] The large language model infers from the prompt statement to obtain consistency detection results for N view models, and also includes:

[0049] When the output consistency check result is inconsistent, backtracking is performed on the inconsistent view model.

[0050] This invention provides a multi-view synthesis method for complex systems based on large models. Figure 1 The consistency detection method, based on the MBSE method, comprehensively considers multi-view design inputs and utilizes a large language model to perform semantic induction and knowledge reorganization on multi-source, complex, and loosely structured modeled view content. Combined with large model prompting engineering techniques, it obtains the corresponding view... Figure 1Consistency testing results significantly enhance the understanding, reasoning, and application capabilities of knowledge in the fields of system architecture and systems engineering, and improve multi-view... Figure 1 The automation, accuracy, and robustness of consistency testing ensure the correctness, completeness, and consistency of the architecture design of complex information systems.

[0051] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0052] Figure 1 Multiview provided in the embodiments of the present invention Figure 1 Flowchart of the biocompatibility testing method;

[0053] Figure 2 Multiview based on DM2 meta-model provided in embodiments of the present invention Figure 1 A schematic diagram illustrating the construction of the consistency detection rule base;

[0054] Figure 3 This is a schematic diagram of the CV-2 capability classification view provided in an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of the CV-4 capability dependency view provided in an embodiment of the present invention;

[0056] Figure 5 The CV-2 capability classification view provided in the embodiments of the present invention Figure 3 Diagram of a tuple set;

[0057] Figure 6 The CV-4 capability-dependent view provided in the embodiments of the present invention Figure 3 Diagram of a tuple set;

[0058] Figure 7 This is a schematic diagram of a subset of CV-2 capability classification view rules provided in an embodiment of the present invention;

[0059] Figure 8 This is a schematic diagram of a subset of CV-4 capability-dependent view rules provided in an embodiment of the present invention;

[0060] Figure 9 This is a schematic diagram of a large-scale detection model based on prompt statements provided in an embodiment of the present invention;

[0061] Figure 10 Multi-view based on a large model provided in the embodiments of the present invention Figure 1 A schematic diagram of the consistency test results. Detailed Implementation

[0062] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0063] As a groundbreaking technology in the field of artificial intelligence, Large Language Models (LLMs) offer new possibilities for solving this problem due to their powerful natural language understanding and generation capabilities. Through training on massive amounts of text data, large models can identify complex semantic relationships and patterns, possessing context learning and reasoning abilities. This capability gives them significant advantages in analyzing textual descriptions of views, comparing information across views, and detecting potential inconsistencies. Although large models have made significant progress in applications in finance, healthcare, and other fields, their application in model systems engineering consistency verification remains a relatively unexplored area, holding enormous development potential.

[0064] Based on this, the present invention proposes a multi-view synthesis method for complex systems based on a large model. Figure 1 The consistency detection method, based on MBSE (Model-Based Systems Engineering), comprehensively addresses multi-view design inputs. It utilizes a large language model to perform semantic induction and knowledge reorganization on multi-source, complex, and loosely structured modeled view content. Combined with large model hinting engineering techniques, it obtains the corresponding view... Figure 1 Consistency testing results significantly enhance the understanding, reasoning, and application capabilities of knowledge in the fields of system architecture and systems engineering, and improve multi-view... Figure 1 Accuracy of contamination testing.

[0065] The Department of Defense Architecture Framework (DoDAF) provides a framework for describing architecture, guiding how to organize and express architectural information for complex systems or enterprises. The DoDAF 2.0 framework includes 52 standardized views across eight categories: capability views, data and information views, project views, operational views, service views, system views, standard views, and panoramic views. Each view is composed of several data elements, which are part of the DoDAF metamodel (DM2) and possess clear semantic definitions and structural constraints, ensuring information consistency and reusability. These views and their data models support architectural expression at multiple levels, from strategic and tactical to technical, and can meet the diverse system architecture requirements of different stakeholders based on their roles and concerns.

[0066] In Model-Based Systems Engineering (MBSE) practice, SysML can be used to create detailed system design schemes, while DoDAF can be used to build high-level architectural views to ensure that the design schemes meet overall strategic goals and architectural requirements. DoDAF focuses on architectural-level descriptions, while SysML focuses on detailed system modeling.

[0067] The following section uses consistency checks of DoDAF model views and SysML model views as examples to illustrate the process.

[0068] like Figure 1 As shown, this invention provides a comprehensive multi-view model for complex systems based on a large model. Figure 1 Inconsistency testing methods include:

[0069] Step 1: Determine the N view models to be tested, construct a structured initial view specification based on the determined view models, and use a large language model to perform semantic induction and knowledge reorganization on the initial view specification to build a rule base;

[0070] (1) Construct the initial view specification.

[0071] Specifically, following the system engineering design methodology of MBSE (Model-Based Systems Engineering), and combining the relevant design requirements of DoDAF and SysML, a structured initial view specification is constructed based on the DoDAF metamodel (DM2) and the SysML view specification.

[0072] 1) The DoDAF metamodel (DM2) describes the system data elements (such as systems, capabilities, tasks, interfaces, information, etc.) available in various views in the DoDAF architecture and the types of relationships that can be established (such as inclusion relationships, dependency relationships, calling relationships, mapping relationships, etc.).

[0073] 2) The SysML view specification details the modeling objectives, semantic composition, element classification, and modeling methods for each view in SysML, and clearly defines the information boundaries and modeling focus of each view.

[0074] (2) A rule base is formed by analyzing and refining large models.

[0075] After completing the construction of the DM2 metamodel and SysML view specifications, Large Language Models (LLMs) are used to semantically summarize and reorganize the multi-source, complex, and loosely structured modeling content. This combines the systematic data structure of DoDAF with the system modeling capabilities of SysML to construct a rule base that integrates two dimensions. Specifically:

[0076] 1) Among them, through in-depth analysis of the semantic relationships, structural mappings and usage patterns of modeling elements between views, the model can automatically extract a set of consistency constraint rules with universality and interpretability.

[0077] 2) Each rule is based on the "relation-view pair", which is formally represented as (relation A, view B, rule R). "Relation A" represents the semantic or structural association between modeling elements, "view B" refers to the model view type to which the rule applies, and "rule R" describes the consistency constraints that need to be satisfied under the specific relation and view context.

[0078] 3) The generated rules are represented in vector form using a text encoder, and all vector representations are stored in the database to form a rule base for easy retrieval later.

[0079] Step 2: Normalize and export the data from the N view models according to the structured XML file, and use regular expression matching to parse the normalized XML file to form a set of triples;

[0080] (1) Exporting XML files of multi-view input models.

[0081] Using DoDAF and SysML view modeling tools, the data in the multi-view model is exported in a structured XML file format. The generated XML file will contain complete project data information, including nodes, relationships, attributes, and view structure. After exporting the XML, nodes with special namespaces (such as reserved information) are deleted, and then XML processing tools are used for formatting, alignment, line breaks, and other processing.

[0082] (2) The normalized and processed XML file is parsed using regular expression matching to form a set of triples.

[0083] 1) First, construct the corresponding tree structure to extract key model element information, including nodes, packages, classes, relationships, and their attributes. Based on the model layer data, the feature IDs and related attribute information of each node and relationship can be identified.

[0084] 2) Subsequently, the graphical information module associated with the view is further parsed to obtain structural descriptions such as the connection relationships between nodes, where the connection relationships will be in the form of (Entity A's ID, Relationship ID, Entity B's ID).

[0085] 3) Combining model layer data and graphical structure information, the entity relationships in each view are systematically reconstructed. That is, based on the ID representation in the graphical information, the relevant element IDs in the model layer data are found, and the ID representations in the original connection relationship representations are replaced with the specific names of nodes and relationships. Finally, a formalized set of triples is generated, such as <entity A, relationship B, entity C>. The triple set of views is then classified according to the view name.

[0086] Step 3: Perform rule retrieval on the set of triples using the rule base to obtain the corresponding rule subset;

[0087] Step 4: Obtain the prompt template, construct the complete prompt statement based on the prompt template and rule subset, input the prompt statement into the large language model, the large language model performs reasoning on the prompt statement, and obtain the consistency detection results of N view models;

[0088] Where N>1.

[0089] This embodiment is based on the DoDAF 2.0 framework, and combines its core metamodel (i.e., DoDAF MetaModel, DM2) and standard architecture view specifications to systematically construct a set of consistency rules between views. This rule system aims to standardize and constrain the expression of relationships between various architecture views, thereby effectively ensuring the consistency of the architecture model in the process of multi-view collaborative modeling.

[0090] Figure 2 Multiview based on DM2 meta-model Figure 1This diagram illustrates the construction of a consistency detection rule base. The overall view (AV) provides the overall background and basic assumptions of the architecture, laying the foundation for the development of subsequent views. The capability view (CV) plays a driving role, guiding the construction of the operational view (OV), and subsequently driving the design of service functions in the service view (SvcV). The operational view depicts the organizational structure, operational scenarios, and key activities, responds to the capability requirements proposed in the capability view, and proposes service support requirements to ensure the achievement of operational missions. The service view (SvcV) undertakes the specific implementation tasks of capabilities, ensuring that its functions support the efficient operation of operational activities. This view forms a mapping relationship with the system view (SV), ensuring that system entities can support service components and their interaction mechanisms, achieving semantic consistency and interface collaboration between services and the system. The standard view (StdV), as the foundation of the architectural technical specifications, uniformly defines the technical standards, protocols, and constraints required for developing various views, ensuring consistency of the architecture at all levels and stages. The project view (PV) provides project-based support for capability and system implementation paths, providing a basis for the implementation of the capability view and system view through time planning and resource allocation. Data and Information Views (DIVs) integrate and model the data structures used or generated by all views, ensuring consistency, availability, and sharing of information, thereby supporting cross-view integration of architectural content.

[0091] After the rule base is built, the data in the view is exported into a structured format, and its content is parsed to extract key entity relationship triples. Taking air rescue as an example, its CV-2 capability view and CV-4 capability dependency view are as follows: Figure 3 and Figure 4 As shown, the extracted sets of triples are respectively as follows: Figure 5 and Figure 6 As shown.

[0092] Based on the extracted entity relation triples and their corresponding view types, matching rule expressions are retrieved from the rule base. This matching process efficiently filters a subset of rules most relevant to the current view content for each triple from the rule base. These rules explicitly define the semantic constraints and modeling criteria that relevant elements in a specific view should follow. Examples of rule subsets are shown below. Figure 7 and Figure 8 As shown.

[0093] After extracting structured triples (e.g., <entity A, relation R, entity B>) from the XML file of the view to be verified and retrieving matching consistency rules, this key information is integrated into a prompt template to construct complete prompt statements, which are then input into the large language model. Furthermore, different corresponding prompt templates are designed for different categories of consistency rules to enhance the generalization performance and specialized processing capabilities of the large model detection. An example of a large model detection based on prompt statements is shown below. Figure 9 As shown.

[0094] By embedding rule content, triplet data, and explicit question instructions in the prompts, the system effectively guides the model to focus on the logical analysis of specific relationships, thereby improving the accuracy and interpretability of judgments. When inconsistencies occur in multi-view data, the system returns inconsistent output results, including explanations and descriptions of the inconsistencies, and supports backtracking of inconsistent views. This is based on a large model with multiple views. Figure 1 Example of consistency test results: Figure 10 As shown in the output, there is an inconsistency between the two views, CV-2 and CV-4. In CV-2, "Aviation Rescue Capability" includes "Information Acquisition Capability", while in CV-4, "Command and Decision-Making Capability" includes "Information Acquisition Capability", indicating an inconsistency in the relationship.

[0095] This embodiment takes the DoDAF model view of complex information system architecture design and the SysML model view of system design as inputs. Following a verification method combining "static, dynamic, and experimental" approaches, it conducts multi-view verification from three aspects: static consistency verification, dynamic simulation consistency verification, and virtual-real linkage experimental consistency verification. Figure 1 Consistency verification.

[0096] This embodiment provides a comprehensive multi-view approach for complex systems based on a large model. Figure 1 The consistency detection method, based on MBSE, integrates multi-view design inputs from DoDAF and SysML. It utilizes a large language model to perform semantic induction and knowledge reorganization on multi-source, complex, and loosely structured modeled view content, and combines large model hinting engineering techniques to obtain the corresponding view... Figure 1 Consistency testing results significantly enhance the understanding, reasoning, and application capabilities of knowledge in the fields of system architecture and systems engineering, and improve multi-view... Figure 1 The automation, accuracy, and robustness of consistency testing ensure the correctness, completeness, and consistency of the architecture design of complex information systems.

[0097] Optionally, in some possible implementations, a large language model is used to perform semantic induction and knowledge reorganization on the initial view specification to construct a rule base, specifically including:

[0098] The large language model parses the initial view specification, analyzes the semantic relationships, structural mappings and usage patterns of modeling elements among N view models, and extracts a set of consistency constraint rules with universality and interpretability.

[0099] Each consistency constraint rule is centered around a relation-view pair and is represented as relation-view-rule. Here, the relation describes the semantic or structural relationship between modeling elements, the view describes the view type of the view model to which the consistency constraint rule applies, and the rule describes the consistency constraints that the relation and view must satisfy.

[0100] The generated consistency constraint rules are represented in vector form using a text encoder, and all vectors are stored in the database to form a rule base.

[0101] Optionally, in some possible implementations, after normalizing and exporting the data from the N view models according to a structured XML file, the method further includes:

[0102] Unnecessary namespace nodes are removed from the exported XML file, and then the XML processing tool is used for formatting, alignment, and line breaks.

[0103] Optionally, in some possible implementations, the normalized exported XML file is parsed using regular expression matching to form a set of triples, specifically including:

[0104] Construct a tree structure corresponding to the XML file, extract model element information, including: nodes, packages, classes, relationships and their attributes, and based on the model layer data, identify the feature IDs and related attribute information of each node and relationship;

[0105] Parse the graphical structure information associated with the view to obtain a structural description of the connection relationships between nodes, wherein the connection relationship is in the form of first entity ID - relationship ID - second entity ID;

[0106] By combining model layer data and graphical structure information, the entity relationships existing in each view model are systematically reconstructed, and finally a formalized set of triples is generated.

[0107] Optionally, in some possible implementations, the entity relationships existing in each view model are systematically reconstructed, specifically including:

[0108] Based on the ID representation in the graphical information, find the relevant element ID in the model layer data, and replace the ID representation in the original connection relationship representation with the specific names of the nodes and relationships.

[0109] Optionally, in some possible implementations, rule retrieval is performed on the set of triples using a rule base to obtain the corresponding rule subset, specifically including:

[0110] By using the view name and relation name corresponding to the triple set, a text encoder is used to obtain the corresponding embedded code. The embedded code is then used to retrieve the rule base to obtain the corresponding preliminary rule subset.

[0111] The rule with the highest similarity in the initial rule subset is determined and returned as the rule for the current set of triples;

[0112] The set returned by all triples is a subset of the rules that the view needs to follow.

[0113] Alternatively, in some possible implementations, the rule base is retrieved according to the following formula:

[0114]

[0115] Where Similarity represents similarity, A represents the embedded encoding of the triple set, and B represents the embedded representation of the rules in the rule base.

[0116] Optionally, in some possible implementations, the prompt template includes:

[0117] Task instructions: Describe the consistency detection task that the large language model needs to complete;

[0118] Contextual information; describes the background content, including input and output examples;

[0119] Input data: The extracted set of triples and the corresponding rule subsets are embedded into the prompt content;

[0120] Output requirements: Guide the large language model to generate response results that conform to the expected format.

[0121] After parsing the set of structured triples and retrieving the matching subset of consistency rules, this key information is integrated into the prompt template to construct a complete prompt statement.

[0122] Inputting the prompts into the large model allows it to leverage its powerful reasoning capabilities to gain multi-view insights. Figure 1 Consistency test results.

[0123] Optionally, in some possible implementations, the large language model performs reasoning on the prompt statement to obtain consistency detection results for N view models, specifically including:

[0124] The large language model infers from the prompt statement. When the data analysis of the N view models of the prompt statement is consistent, the output consistency detection result is consistent.

[0125] When the data analysis of the N view models in the prompt statement is inconsistent, the consistency check result output is inconsistent, and the output result includes an explanation and description of the inconsistent content.

[0126] Optionally, in some possible implementations, the large language model infers the prompt statement to obtain consistency detection results for N view models, and further includes:

[0127] When the output consistency check result is inconsistent, backtracking is performed on the inconsistent view model.

[0128] Optionally, some possible implementations may include all or part of the above-described implementations.

[0129] It should be understood that in the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this description, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate different embodiments or examples described in this specification, as well as some features of different embodiments or examples.

[0130] Of course, those skilled in the art can make various corresponding changes and modifications based on the present invention without departing from its spirit and essence, but such changes and modifications should all fall within the protection scope of the claims of the present invention.

Claims

1. A method for consistency detection of multiple views in a complex system based on a large model, characterized in that, include: N view models to be detected are identified. A structured initial view specification is constructed based on the identified view models. A large language model is used to perform semantic induction and knowledge reorganization on the initial view specification to construct a rule base. The data from N view models are normalized and exported according to the structured XML file. The normalized XML file is parsed using regular expression matching to form a set of triples. The rule set of triples is retrieved by the rule base to obtain the corresponding rule subset; Obtain a prompt template, construct a complete prompt statement based on the prompt template and the rule subset, input the prompt statement into the large language model, and the large language model performs reasoning on the prompt statement to obtain the consistency detection results of N view models; Where N > 1.

2. The method for detecting consistency of multiple views in complex systems based on large models according to claim 1, characterized in that, The initial view specification is semantically summarized and knowledge reorganized using a large language model to construct a rule base, specifically including: The large language model parses the initial view specification, analyzes the semantic association, structural mapping and usage patterns of the modeling elements among the N view models, and extracts a set of consistency constraint rules with universality and interpretability. Each consistency constraint rule is centered around a relation-view pair and is represented as relation-view-rule. The relation describes the semantic or structural association between modeling elements, the view describes the view type of the view model to which the consistency constraint rule applies, and the rule describes the consistency constraints that the relation and the view must satisfy. The generated consistency constraint rules are represented in vector form using a text encoder, and all vectors are stored in the database to form a rule base.

3. The method for detecting consistency of multiple views in complex systems based on large models according to claim 1, characterized in that, After normalizing and exporting the data from N view models according to the structured XML file format, it also includes: Unnecessary namespace nodes are removed from the exported XML file, and then the XML processing tool is used for formatting, alignment, and line breaks.

4. The method for detecting consistency of multiple views in complex systems based on large models according to claim 1, characterized in that, The normalized exported XML file is parsed using regular expression matching to form a set of triples, specifically including: Construct a tree structure corresponding to the XML file, extract model element information, including: nodes, packages, classes, relationships and their attributes, and identify the feature IDs and related attribute information of each node and relationship based on the model layer data; Parse the graphical structure information associated with the view to obtain a structural description of the connection relationships between nodes, wherein the connection relationship is in the form of first entity ID - relationship ID - second entity ID; By combining model layer data and graphical structure information, the entity relationships existing in each view model are systematically reconstructed, ultimately generating a formalized set of triples.

5. The method for detecting consistency of multiple views in complex systems based on large models according to claim 4, characterized in that, The system refactors the entity relationships existing in each view model, specifically including: Based on the ID representation in the graphical information, find the relevant element ID in the model layer data, and replace the ID representation in the original connection relationship representation with the specific names of the nodes and relationships.

6. The method for detecting consistency of multiple views in complex systems based on large models according to claim 1, characterized in that, The rule base is used to perform rule retrieval on the set of triples to obtain the corresponding rule subset, specifically including: By using the view name and relation name corresponding to the triple set, a text encoder is used to obtain the corresponding embedded code. The embedded code is then used to retrieve the rule base to obtain the corresponding preliminary rule subset. The rule with the highest similarity in the initial rule subset is determined and returned as the rule for the current set of triples; The set returned by all triples is a subset of the rules that the view needs to follow.

7. The method for detecting consistency of multiple views in complex systems based on large models according to claim 6, characterized in that, The rule base is retrieved using the following formula: Where Similarity represents similarity, A represents the embedded encoding of the triple set, and B represents the embedded representation of the rules in the rule base.

8. The method for detecting consistency of multiple views in complex systems based on large models according to claim 1, characterized in that, The prompt templates include: Task instructions: Describe the consistency detection task that the large language model needs to complete; Contextual information: Describes the background content, including input and output examples; Input data: The extracted set of triples and the corresponding rule subsets are embedded into the prompt content; Output requirements: Guide the large language model to generate response results that conform to the expected format.

9. The method for detecting consistency of multiple views in complex systems based on large models according to claim 1, characterized in that, The large language model infers from the prompt statement to obtain consistency detection results for N view models, specifically including: The large language model infers the prompt statement, and when the data analysis of the N view models of the prompt statement is consistent, the output consistency detection result is consistent. When the data analysis of the N view models in the prompt statement is inconsistent, the consistency check result output is inconsistent, and the output result includes an explanation and description of the inconsistent content.

10. The method for detecting consistency of multiple views in complex systems based on large models according to claim 9, characterized in that, The large language model infers from the prompt statement to obtain consistency detection results for N view models, and also includes: When the output consistency check result is inconsistent, backtracking is performed on the inconsistent view model.

Images