Traffic situation knowledge incremental evolution method and system based on ontology structure constraint

Abstract

This invention relates to a method and system for incremental evolution of traffic situation knowledge based on ontology structure constraints, comprising: obtaining an initial complete ontology; performing structured transcription and chapter-level organization of traffic-related literature corpora, calling a large language model to evaluate modeling value, and screening effective corpus units; extracting core class levels and key attributes from the initial complete ontology, generating a simplified ontology structure, and inputting it together with the effective corpus units into the large language model, extracting candidate ontology elements under chain-guided extraction; calculating the semantic similarity within the candidate elements for deduplication and merging, then comparing it with the existing ontology in the large language model to identify elements to be updated; inputting the elements to be updated together with the initial complete ontology into the large language model, determining inclusion feasibility and assigning structural positions under global constraints, and formally writing them into the generated updated ontology. This invention achieves automated incremental evolution of traffic situation knowledge under simplified and complete two-level ontology structure constraints, ensuring consistency and structural stability in knowledge updates.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and knowledge engineering technology, and in particular to a method and system for incremental evolution of traffic situation knowledge based on ontology structure constraints. Background Technology

[0002] Urban transportation systems are typical complex mega-systems, with rapidly evolving operational states and numerous influencing factors, resulting in highly dynamic and structurally complex traffic situational knowledge. Accurately understanding and effectively organizing this knowledge is a crucial foundation for precise traffic congestion tracing, dynamic travel route planning, and regional collaborative management. Ontology, as a formal knowledge representation model, can clearly define concepts, attributes, relationships, and constraints in the transportation domain, providing an ideal framework for the unified description and structured organization of heterogeneous traffic knowledge.

[0003] However, urban traffic situation knowledge itself is constantly evolving. New traffic indicators, congestion evolution patterns, event types, and dynamic relationships between objects emerge continuously as research deepens and practice develops. Existing ontology construction methods mainly rely on manual editing and updating by domain experts, which is difficult to adapt to the rapid changes and growth in scale of knowledge, and faces the following significant limitations:

[0004] (i) Low efficiency in knowledge acquisition and updating. New knowledge is scattered across a vast amount of research literature, technical reports and business specifications. The process of manually identifying, extracting and converting it into ontological elements is time-consuming and laborious, making it difficult to meet the timeliness requirements of real-time or near-real-time traffic situation analysis.

[0005] (ii) Knowledge consistency is difficult to guarantee. New knowledge often has different levels of abstraction, spatial scale, or semantic granularity. How to maintain its consistency with the existing ontology class hierarchy, attribute constraints, and semantic relationships during the update process, and avoid introducing conflicts or redundancy, is a problem that traditional methods cannot solve systematically.

[0006] (iii) Lack of automated evolution capability. Most existing methods are constructed "once" and lack the ability to automatically complete closed-loop evolution of knowledge screening, element extraction, semantic alignment, structural determination and ontology dynamic update under the continuous input of new corpus, which makes it difficult for the knowledge base to grow in sync with the development of the domain.

[0007] Therefore, a method is needed that can automatically drive the incremental evolution of ontology knowledge using new corpora under the constraints of existing ontology structures, so as to improve the dynamic adaptability, structural consistency and automation level of traffic situation knowledge modeling. Summary of the Invention

[0008] In view of the shortcomings of the prior art, the purpose of this invention is to provide a method and system for incremental evolution of traffic situation knowledge based on ontology structure constraints, so as to solve one or more problems in the prior art.

[0009] To achieve the above objectives, the technical solution of the present invention is as follows:

[0010] The traffic situation knowledge incremental evolution method based on ontology structure constraints includes the following steps:

[0011] S1. Obtain the initial complete ontology: Obtain the ontology structure containing multi-level traffic situation knowledge and use it as the initial complete ontology for the self-evolution process;

[0012] S2. Literature Corpus Screening and Effective Corpus Construction: Obtain literature corpus in the field of transportation and perform structured transcription and chapter hierarchical organization. Call the large language model to perform ontology modeling value assessment on chapter titles, and divide the screened chapters into effective corpus units according to semantic boundaries.

[0013] S3. Feature extraction based on simplified ontology structure constraints: Extract the core class hierarchy, key object attributes and data attributes from the initial complete ontology to generate a simplified ontology structure that retains the sub-extension capabilities of the parent class; use the simplified ontology structure as a prior constraint, input it together with the effective corpus units into the large language model, and use chained guiding prompts to drive the large language model to sequentially perform ontology prior understanding, semantic understanding, candidate recognition, boundary control and type determination;

[0014] S4. Candidate element alignment and deduplication: Calculate the semantic vector similarity within the candidate element set to merge duplicates and generate normalized candidate elements; then call the large language model to compare the normalized candidate elements with existing ontology elements. If they are not fully covered by the existing ontology, they are retained as elements to be updated.

[0015] S5. Incremental update based on complete ontology structure constraints: Input the elements to be updated and the initial complete ontology into the large language model. Under the global constraints of the complete ontology class hierarchy and attribute association, determine the inclusion of the elements to be updated and assign structural positions. Formalize the elements that pass the determination and write them into the initial complete ontology to generate the updated ontology, realizing the incremental evolution of ontology knowledge.

[0016] Furthermore, in step S1, the initial complete ontology is a hierarchical ontology containing six levels, including:

[0017] The time information layer and spatial information layer are used to provide spatiotemporal anchor points;

[0018] The micro-level traffic situation layer and the aggregated traffic situation layer are used to carry operational data;

[0019] The congestion process layer and the general traffic event layer are used to express the dynamic evolution and external disturbances.

[0020] Cross-level relationships are established between the six levels through object properties.

[0021] Furthermore, the judgment rules followed by the large language model in step S2 for corpus selection include:

[0022] Domain Scope Determination Rules: Determine whether the chapter title and level point to the urban traffic situation domain;

[0023] Modeling Clue Recognition Rules: Identify whether the title contains business concept clues;

[0024] Non-modeling content filtering rules: Identify whether the title refers to purely technical or experimental discussions;

[0025] Modeling value comprehensive evaluation rules: For compound titles, determine their semantic focus. If the focus is biased towards traffic situation business knowledge, output the retained conclusion and convert it into effective corpus units.

[0026] Furthermore, in step S3, the large language model extracts elements based on the simplified ontology structure, following the type determination rules:

[0027] Candidate class determination rule: If the concept in the effective corpus unit is dynamic, time-varying, or an operational concept that can be observed or measured under specific spatiotemporal conditions, it is given priority as a candidate class, and a suggested parent class is given based on the simplified structure of the ontology;

[0028] Candidate object attribute determination rules: If it expresses a stable, abstract, and reusable semantic relationship between two traffic knowledge objects, it is given priority as a candidate object attribute, and suggestions on the domain and range are given based on the simplified structure of the ontology.

[0029] Candidate data attribute determination rules: If the expression is a static, stable, or configurable literal attribute that is attached to a certain type of object, it is given priority as a candidate data attribute, and suggestions on the domain and data type are given.

[0030] Furthermore, in step S4, the large language model compares the normalized candidate elements with existing ontology elements, following semantic comparison rules including:

[0031] Semantic equivalence judgment rule: Determine whether the candidate element and the existing ontology element express the same transportation concept, the same relationship, or the same attribute;

[0032] Granularity consistency judgment rule: Determine whether the candidate element and the existing ontology element are at the same abstraction level and the same spatial scale;

[0033] Coverage relationship judgment rule: Determine whether the existing ontology elements have fully expressed the core semantics of the candidate elements;

[0034] Candidate elements are merged into existing ontology elements only if all three rules are met; otherwise, candidate elements are retained as elements to be updated.

[0035] Furthermore, in step S5, the large language model is incrementally updated based on the complete ontology structure, following the following judgment and writing rules:

[0036] Inclusion criteria: First, determine whether the element to be updated has independent modeling value. If its essence is a specific instance, algorithm process, model method, or research task, then output a decision to reject inclusion; if it has a stable semantic boundary and is related to the traffic situation business concept, then continue to perform structural positioning.

[0037] Structural positioning rules include:

[0038] Class element location rule: Select the parent class with the closest semantics and the most stable hierarchy from the complete ontology, and attach the class element to be updated to the parent class through the rdfs:subClassOf relationship;

[0039] Object attribute location rules: Force the domain and value range of the attribute to be updated to be anchored to the available class anchor point of the complete ontology, and establish the association through rdfs:domain and rdfs:range;

[0040] Data attribute location rules: Force the domain of the data attribute to be updated to be anchored to the available class anchor point of the complete ontology, and bind its value type to a valid XSD data type through rdfs:range;

[0041] Update decision and write rules: Based on the results of inclusion judgment and structural positioning, the final update decision is output; when the decision is that it is locatable, the elements to be updated and their structural positions are formalized into OWL semantic descriptions and written into the initial complete ontology; when the decision is questionable or rejected, no writing is performed, and the manual review interface is retained.

[0042] The traffic situation knowledge incremental evolution system based on ontology structure constraints adopts the above-mentioned traffic situation knowledge incremental evolution method based on ontology structure constraints, including:

[0043] The corpus structuring and filtering module is configured to transcribe traffic documents into structured text, call a large language model to evaluate the chapter modeling value, and generate effective corpus units.

[0044] The prior constraint extraction module is configured to extract a simplified ontology structure from the initial complete ontology, and under the prior constraints of the simplified ontology structure, call the large language model to extract structured candidate ontology elements from the effective corpus units.

[0045] The two-level semantic alignment module is configured to perform internal repetitive convergence of candidate ontology elements by calculating vector similarity, and call the large language model to determine the equivalence, granularity and coverage relationship between candidate elements and existing ontology, and output the elements to be updated.

[0046] The structure-aware evolution module is configured to call the large language model to determine the inclusion of elements to be updated and assign precise logical structure positions under the global constraints of the initial complete ontology structure. Finally, it writes these elements into the ontology through formal code to achieve evolutionary updates.

[0047] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for incremental evolution of traffic situation knowledge based on ontology structure constraints.

[0048] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method for incremental evolution of traffic situation knowledge based on ontology structure constraints.

[0049] Compared with the prior art, the beneficial technical effects of the present invention are as follows:

[0050] (i) Achieving Automatic Incremental Evolution of Knowledge: This invention constructs a complete automated chain of "corpus screening - element extraction - semantic alignment - structural update". By introducing a large language model and ontology structure constraints, it can automatically identify, extract and update ontology elements under the drive of continuously input new literature corpus, realizing the dynamic and sustainable evolution of the traffic situation knowledge base and significantly reducing manual maintenance costs.

[0051] (II) Ensuring Structural Consistency in the Evolutionary Process: This invention innovatively proposes a dual guarantee mechanism combining "prior simplified structural constraints" and "posterior complete structural constraints." During the extraction phase, the simplified ontology structure guides the large model to focus on core semantics, avoiding deviations from the existing framework. During the update phase, the complete ontology structure rigorously determines the structural positions of new elements, such as their parent class, domain, and value domain, ensuring global consistency between new knowledge and existing ontology in terms of class hierarchy, attribute associations, and logical constraints, effectively avoiding knowledge conflicts and structural redundancy.

[0052] (III) Enhancing the accuracy and robustness of knowledge acquisition: This invention designs refined chain-like guiding prompts and multi-level judgment rules (such as semantic equivalence, granularity consistency, and coverage relationship judgment). Combining deduplication of semantic vectors within candidate elements with secondary alignment based on a large model, it can effectively handle the differences in terminology, granularity, and context among different documents, accurately identify truly novel elements that have modeling value, and improve the accuracy of knowledge extraction and fusion. Attached Figure Description

[0053] Figure 1 The diagram shows the overall flowchart of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention.

[0054] Figure 2 The diagram illustrates a six-layer structure of the initial complete ontology of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints, according to an embodiment of the present invention.

[0055] Figure 3 The diagram illustrates a comparison between the complete ontology structure and the simplified ontology structure of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention.

[0056] Figure 4 This illustration shows a flowchart of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention, which uses a large language model for chain-like guidance and judgment in corpus screening.

[0057] Figure 5 This illustration shows a flowchart of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention, which extracts ontology elements from a large language model for chain-like guidance and judgment.

[0058] Figure 6 This diagram illustrates the process of internal alignment and normalized candidate element generation for the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention.

[0059] Figure 7 This illustration shows a flowchart of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention, which performs chain-like guidance and judgment on ontology element alignment with a large language model.

[0060] Figure 8 This illustration shows a flowchart of the incremental evolution method for traffic situation knowledge based on ontology structure constraints and the system for chain-guided judgment of ontology element update of the large language model in an embodiment of the present invention.

[0061] Figure 9 This illustration shows a structured candidate ontology element example generated after ontology element extraction from an effective corpus unit of the traffic situation knowledge incremental evolution method and system based on ontology structure constraints according to an embodiment of the present invention.

[0062] Figure 10 This illustration shows a structured update suggestion generated after an ontology element update determination for a traffic situation knowledge incremental evolution method and system based on ontology structural constraints, according to an embodiment of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the following detailed description of the traffic situation knowledge incremental evolution method and system based on ontological structural constraints proposed in this invention, in conjunction with the accompanying drawings and specific embodiments, will provide a more detailed explanation. The advantages and features of this invention will become clearer from the following description. It should be noted that the accompanying drawings are in a very simplified form and use non-precise proportions, used only to facilitate and clearly illustrate the purpose of the embodiments of this invention. Please refer to the accompanying drawings to make the objectives, features, and advantages of this invention more apparent and understandable. It should be understood that the structures, proportions, sizes, etc., depicted in the accompanying drawings are only for illustrative purposes to aid those skilled in the art and are not intended to limit the implementation conditions of this invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to the size, without affecting the effects and objectives achieved by this invention, should still fall within the scope of the technical content disclosed in this invention.

[0064] Please see Figure 1 The traffic situation knowledge incremental evolution method based on ontology structure constraints includes the following steps:

[0065] Step S1: Obtain the initial complete ontology.

[0066] Obtain an ontology structure containing multi-level traffic situation knowledge, and use it as the initial complete ontology for the self-evolution process. The initial complete ontology is as follows: Figure 2 The layered ontology shown comprises six levels, specifically: a time information layer and a spatial information layer for providing spatiotemporal anchors; a micro-traffic situation layer and an aggregated traffic situation layer for carrying operational data; a congestion process layer and a general traffic event layer for expressing dynamic evolution and external disturbances; and cross-level associations between the six levels are established through object attributes.

[0067] In this embodiment, the time information layer uses the time information class as the superclass, with subclasses "time slice" and "time period". The spatial information layer uses the spatial information class as the superclass, with four key subclasses: region, road, road segment, and intersection. The micro-traffic situation layer sets up road segment traffic status and intersection traffic status classes, and associates them with road segment traffic indicators and intersection traffic indicators, respectively. The aggregated traffic situation layer sets up road traffic status and regional traffic status classes, and associates them with road traffic indicators and regional traffic indicators, respectively. The congestion process layer sets up congestion event and congestion evolution status classes. The general traffic event layer sets up a general traffic event class, which includes subclasses such as traffic accidents, road construction, traffic control, and severe weather.

[0068] Step S2: Literature corpus screening and effective corpus construction.

[0069] We acquire literature corpora in the field of transportation and perform structured transcription and chapter-level organization. We call a large language model to perform ontology modeling value assessment on chapter titles and divide the selected chapters into effective corpus units according to semantic boundaries.

[0070] In this embodiment, specifically:

[0071] Step S2-1, Text Recognition and Structure Transcribing: Using Optical Character Recognition (OCR) technology, the text content in the original document is extracted into processable text, and the heading levels and paragraph boundaries are sorted out according to the typesetting characteristics, and the headings and body text are organized into continuous hierarchical text.

[0072] Step S2-2, Corpus Screening and Large Language Model Invocation: The chapter titles and their hierarchical information are input to the corpus screening large language model. This model follows the following... Figure 4 The chain-like guidance and judgment process shown executes the steps of determining the domain scope, judging the semantic orientation of the title, identifying the orientation of non-modeling content, and evaluating relevance and modeling value in sequence.

[0073] The decision rules followed by the model include:

[0074] Domain scope determination rules: Determine whether the chapter title and level point to the urban traffic situation domain. This corresponds to the domain scope determination step, which aims to quickly filter out cross-domain noisy corpora and narrow the scope of subsequent semantic analysis.

[0075] Modeling Clue Identification Rules: Identify whether the title contains business concept clues such as traffic objects, states, indicators, and events. Corresponding to the title semantic orientation judgment step, and based on the domain scope determination step, further analyze the title to identify whether it contains business concept clues such as traffic objects, operational states, traffic indicators, event processes, and structural relationships, in order to locate the core business elements in the title that have potential value for ontology modeling.

[0076] Non-modeling content filtering rules: Identify whether the title points to purely technical or experimental discussions such as data processing, algorithm implementation, or experimental parameter settings. Correspondingly, identify the steps indicated by the non-modeling content to extract non-knowledge-based distracting information.

[0077] The comprehensive evaluation rule for modeling value is as follows: For composite titles, determine their semantic focus. If the focus leans towards traffic situation business knowledge, output a retained conclusion. Corresponding to the steps of evaluating relevance and modeling value, and combining the identification results of the steps of determining the semantic orientation of the title and identifying the orientation of non-modeling content, further determine whether the semantic focus of the composite title leans towards traffic situation business knowledge. This avoids the model making mechanical judgments based solely on a single keyword and ensures the accuracy of complex corpus screening.

[0078] Step S2-3: Construction of Effective Corpus Units: For the chapters that the model outputs as "retained," their main text content is backfilled, and the text is segmented into relatively focused text fragments according to semantic boundaries to form effective corpus units. Each unit includes source information such as document number, chapter title, level, and main text.

[0079] Step S3: Extracting elements based on simplified structural constraints of ontology.

[0080] like Figure 3 As shown, the core class hierarchy, key object attributes, and data attributes in the initial complete ontology are extracted to generate a simplified ontology structure that retains the sub-extension capabilities of the parent class. The simplified ontology structure is used as a priori constraint and input into the large language model along with effective corpus units. Chained guiding prompts are used to drive the large language model to sequentially perform ontology prior understanding, semantic understanding, candidate recognition, boundary control, and type determination.

[0081] In this embodiment, specifically:

[0082] Step S3-1: Construct a simplified ontology structure: Extract core parent classes with lower-level extension capabilities (such as "Spatial Information Class", "Traffic Status Class", "Traffic Indicator Class") and key object attributes with clear domains and value ranges (such as "Describes Spatial Objects" and "Aggregation Originates From") from the initial complete ontology to form a simplified ontology structure in Turtle format.

[0083] Step S3-2, Ontology Feature Extraction and Large Language Model Invocation: Input the simplified ontology structure and effective corpus units into the model. The model follows the following... Figure 5 The chain-guided process shown in the figure is used to process the data in sequence, including understanding the prior ontology, understanding the semantics of effective corpus units, identifying candidate elements, performing boundary control, type-specific determination, and providing preliminary structure suggestions.

[0084] The type determination rules followed by the model include:

[0085] Candidate category determination rules: If a concept in a valid corpus unit is dynamic, time-varying, or an operational concept that can be observed or measured under specific spatiotemporal conditions, it is given priority as a candidate class, and a suggested parent class is provided based on the simplified ontology structure. For example, from the corpus "Road segment saturation, travel time ratio, and travel speed were selected as three secondary indicators for traffic situation evaluation in this paper," "Road segment saturation" can be determined as a candidate class, and its parent class is suggested to be "Road segment traffic indicators." The corresponding type-specific determination and preliminary structure suggestion steps, based on the boundary control steps to exclude expressions unsuitable for entering the ontology schema layer, and combined with the core class hierarchy obtained in the prior ontology understanding steps, determine independent semantic concepts and provide parent class suggestions to achieve hierarchical positioning of core business concepts.

[0086] Candidate object attribute determination rules: If an attribute expresses a stable, abstract, and reusable semantic relationship between two traffic knowledge objects, it is prioritized as a candidate object attribute, and suggestions for its domain and value range are provided. For example, from the corpus "calculate road segment saturation based on traffic flow", "calculated from road segment traffic flow" can be determined as a candidate object attribute, with a suggested domain of "road segment saturation" and a suggested value range of "road segment traffic flow". The corresponding type-specific determination and preliminary structure suggestion steps, based on the boundary control steps to filter and eliminate interfering expressions, and based on understanding the key attribute relationships and structural constraints in the ontology prior steps, determine the semantic relationship between two knowledge objects and provide suggestions for their domain and value range to establish a stable entity association structure in the ontology.

[0087] Candidate data attribute determination rules: If the expression is a static, stable, or configurable literal attribute that depends on a certain type of object, it is given priority as a candidate data attribute, and suggestions for its domain and data type are provided. The corresponding type-specific determination and preliminary structure suggestion steps, after filtering in the boundary control step, combine the attribute and structural constraint specifications from the ontology prior step to determine the static, stable, or configurable characteristics that depend on a specific object and provide data type suggestions to improve the underlying static attribute constraints of the ontology.

[0088] Step S3-3, Structured Output: The model outputs structured candidate ontology elements in strict JSON format. A typical output is as follows: Figure 9 As shown, it includes element names, types, definitions, original evidence, and structural suggestions.

[0089] Step S4: Align and deduplication of candidate elements.

[0090] The semantic vector similarity within the candidate element set is calculated to merge duplicates and generate normalized candidate elements. Then, the large language model is called to compare the normalized candidate elements with existing ontology elements. If the normalized candidate elements are not fully covered by the existing ontology, they are retained as elements to be updated.

[0091] In this embodiment, specifically:

[0092] Step S4-1, Internal Alignment of Candidate Elements: The BGE-M3 embedding model is used to map the text representations of candidate elements into semantic vectors, and cosine similarity is calculated. A preset similarity threshold (e.g., 0.85) is set, and candidate elements with similarity values ​​higher than the threshold are grouped into the same candidate cluster. Figure 6 As shown, different expressions such as "segment saturation" and "segmentsaturation" in the same candidate cluster are integrated into a standardized candidate element, which includes the main name, alias set, fusion definition and multi-party evidence.

[0093] Step S4-2, Aligning Candidate Elements with Existing Ontologies: The normalized candidate elements and candidate targets from existing ontologies (recalled through similarity) are jointly input into the ontology element alignment large language model. The model follows the following... Figure 7 The chain-guided process shown determines semantic equivalence, granularity consistency, and coverage relationships.

[0094] The semantic alignment rules followed by the model include:

[0095] Semantic equivalence judgment rule: Determine whether candidate elements and existing ontology elements express the same transportation concept, the same relationship, or the same attribute. Corresponding to the semantic equivalence judgment step, based on a deep understanding of the contextual semantics of both parties (names, definitions, aliases, scope of expression, etc.) in the steps of understanding the candidate elements to be aligned and understanding the existing ontology candidate targets, compare whether the conceptual and attribute features of the two have the same semantics. This is used to identify equivalence targets with potential alignment possibilities from the perspective of connotation.

[0096] Granularity consistency judgment rule: Determine whether both are at the same abstract level and spatial scale. "Road segment average speed" and "regional average speed" have inconsistent granularity and cannot be merged. Corresponding to the steps of judging granularity consistency and expression scope, based on the semantic equivalence judgment step, further compare the two in terms of abstract level, spatial scale, business scope, and structural position. This is to accurately identify granularity differences such as macro and micro, higher and lower levels, avoiding logical confusion and erroneous merging at different granularity levels.

[0097] Coverage relationship judgment rule: Determine whether the existing ontology elements have fully expressed the core semantics of the candidate elements. Corresponding to the coverage relationship judgment step, on the premise that the granularity consistency judgment and expression range judgment steps are passed, further evaluate the capacity and expression completeness of the existing ontology elements to accommodate the core semantics of the candidate elements, so as to provide a precise coverage basis for the final alignment judgment.

[0098] Candidate elements are merged into existing ontology elements only if all three of the above rules are met; otherwise, candidate elements are retained as elements to be updated.

[0099] Step S5: Incremental update based on complete ontology structure constraints.

[0100] The elements to be updated are input into the large language model along with the initial complete ontology. Under the global constraints of the complete ontology class hierarchy and attribute association, the inclusion of the elements to be updated is determined and structural positions are assigned. The elements that pass the determination are formally written into the initial complete ontology to generate the updated ontology, thereby realizing the incremental evolution of ontology knowledge.

[0101] In this embodiment, specifically:

[0102] Step S5-1, Ontology Feature Update: Large Language Model Call: The model follows the following... Figure 8 The chain-guided process shown sequentially understands the complete ontology structure, understands the semantics of the elements to be updated, determines inclusion feasibility, and determines the structural position. In this embodiment, class elements and attribute elements are processed separately, and the determination of attribute elements depends on the already approved class element anchor points.

[0103] The decision and write rules followed by the model include:

[0104] Inclusion Judgment Rules: First, determine whether the element to be updated has independent modeling value. If its essence is a specific instance, algorithm process, or research task such as "specific intersection name," "KNN algorithm," or "simulation experiment process," then output REJECTED (reject inclusion). If it has stable semantic boundaries and is related to traffic situation business concepts, then continue with structural positioning. Corresponding to the inclusion judgment step, based on understanding the existing ontology structure steps and the semantic steps of the element to be updated, analyze whether the element to be updated has independent modeling value and ontology layer compatibility. This is to fundamentally filter out interfering content such as specific instances and algorithms that do not conform to the ontology model layer, ensuring that only basic business elements with stable semantic boundaries enter the subsequent structural positioning stage.

[0105] Structural positioning rules include:

[0106] Element location rules: Select the parent class with the closest semantics and the most stable hierarchy from the complete ontology. For example, for the class to be updated, "intersection lane-level queue length," the model determines that its semantics are a refinement of intersection queue length, and suggests "intersection queue length" as the parent class, rather than the broader "intersection traffic indicators." Correspondingly, in the structural location determination step, after the inclusion decision is passed, the existing ontology structure is combined to lock the closest reasonable parent class node in the complete ontology.

[0107] Object attribute location rule: Forces the domain and value range of the object attribute to be updated to be anchored to the available class anchor point of the complete ontology. Corresponding to the structural location judgment step, after the inclusion judgment passes, it is used to ensure that the domain and value range of the object attribute are legally bound to the existing ontology class, establishing a stable entity association.

[0108] Data attribute location rules: The domain of the data attribute to be updated is forced to be anchored to an available class anchor point in the complete ontology, and its value type is bound to a valid XSD data type via rdfs:range. Corresponding to the structural location determination step, after the inclusion determination passes, this is used to constrain the subject and numerical format of static attributes, thus improving the semantic constraints at the underlying level of the ontology.

[0109] Update Decision and Write Rules: Output the final decision based on the comprehensive judgment results. Corresponding to the confidence step of the judgment result and the generation of structured output steps, under the premise that the location suggestion is output in the structural location judgment step, the confidence assessment is completed by comprehensively considering the location clarity and the sufficiency of evidence support. This results in the unified generation and output of a structured update judgment result containing the decision, confidence level, location, and reason, providing a high-confidence decision basis for subsequent formal construction and writing. When the decision is POSITIONED (locatable), a structured update suggestion is generated, such as... Figure 10 As shown, it includes the decision, confidence level, suggested parent class, and reason. Subsequently, the elements to be updated and their structural positions are formalized into OWL semantic descriptions.

[0110] Step S5-2, Formal Construction and Writing: The checked features to be included are converted into OWL / Turtle format statements. For class features, an owl:Class declaration and rdfs:subClassOf relation are generated; for object attributes, an owl:ObjectProperty declaration and rdfs:domain and rdfs:range are generated; for data attributes, an owl:DatatypeProperty declaration and corresponding domain and value type constraints are generated. These statements are appended to the initial complete ontology file to form the updated ontology.

[0111] The traffic situation knowledge incremental evolution system based on ontology structure constraints adopts the above-mentioned traffic situation knowledge incremental evolution method based on ontology structure constraints, including:

[0112] The corpus structuring and filtering module is configured to transcribe traffic documents into structured text, call large language models to evaluate the chapter modeling value, and generate effective corpus units.

[0113] The prior constraint extraction module is configured to extract a simplified ontology structure from the initial complete ontology, and under the prior constraints of this structure, call the large language model to extract structured candidate ontology elements from the effective corpus units.

[0114] The two-level semantic alignment module is configured to perform internal repetitive convergence of candidate ontology elements by calculating vector similarity, and call the large language model to determine the equivalence, granularity and coverage relationship between candidate elements and existing ontology, and output the elements to be updated.

[0115] The structure-aware evolution module is configured to call the large language model to determine the inclusion of elements to be updated and assign precise logical structure positions under the global constraints of the initial complete ontology structure. Finally, it writes these elements into the ontology through formal code to achieve evolutionary updates.

[0116] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for incremental evolution of traffic situation knowledge based on ontology structure constraints.

[0117] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described traffic situation knowledge incremental evolution method based on ontology structure constraints.

[0118] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A traffic situation knowledge incremental evolution method based on ontology structure constraints, characterized by: Includes the following steps: S1. Obtain the initial complete ontology: Obtain the ontology structure containing multi-level traffic situation knowledge and use it as the initial complete ontology for the self-evolution process; S2. Literature Corpus Screening and Effective Corpus Construction: Obtain literature corpus in the field of transportation and perform structured transcription and chapter hierarchical organization. Call the large language model to perform ontology modeling value assessment on chapter titles, and divide the screened chapters into effective corpus units according to semantic boundaries. S3. Feature extraction based on simplified ontology structure constraints: Extract the core class hierarchy, key object attributes and data attributes from the initial complete ontology to generate a simplified ontology structure that retains the sub-extension capabilities of the parent class; use the simplified ontology structure as a prior constraint, input it together with the effective corpus units into the large language model, and use chained guiding prompts to drive the large language model to sequentially perform ontology prior understanding, semantic understanding, candidate recognition, boundary control and type determination; S4. Candidate element alignment and deduplication: Calculate the semantic vector similarity within the candidate element set to merge duplicates and generate normalized candidate elements; then call the large language model to compare the normalized candidate elements with existing ontology elements. If they are not fully covered by the existing ontology, they are retained as elements to be updated. S5. Incremental update based on complete ontology structure constraints: Input the elements to be updated and the initial complete ontology into the large language model. Under the global constraints of the complete ontology class hierarchy and attribute association, determine the inclusion of the elements to be updated and assign structural positions. Formalize the elements that pass the determination and write them into the initial complete ontology to generate the updated ontology, realizing the incremental evolution of ontology knowledge.

2. The traffic situation knowledge incremental evolution method based on ontology structure constraints as described in claim 1, characterized in that: In step S1, the initial complete ontology is a hierarchical ontology containing six levels, including: The time information layer and spatial information layer are used to provide spatiotemporal anchor points; The micro-level traffic situation layer and the aggregated traffic situation layer are used to carry operational data; The congestion process layer and the general traffic event layer are used to express the dynamic evolution and external disturbances. Cross-level relationships are established between the six levels through object properties.

3. The traffic situation knowledge incremental evolution method based on ontology structure constraints as described in claim 1, characterized in that: The judgment rules followed by the large language model in step S2 for corpus selection include: Domain Scope Determination Rules: Determine whether the chapter title and level point to the urban traffic situation domain; Modeling Clue Recognition Rules: Identify whether the title contains business concept clues; Non-modeling content filtering rules: Identify whether the title refers to purely technical or experimental discussions; Modeling value comprehensive evaluation rules: For compound titles, determine their semantic focus. If the focus is biased towards traffic situation business knowledge, output the retained conclusion and convert it into effective corpus units.

4. The traffic situation knowledge incremental evolution method based on ontology structure constraints as described in claim 1, characterized in that: In step S3, the large language model extracts elements based on a simplified ontology structure, following the following type determination rules: Candidate class determination rule: If the concept in the effective corpus unit is dynamic, time-varying, or an operational concept that can be observed or measured under specific spatiotemporal conditions, it is given priority as a candidate class, and a suggested parent class is given based on the simplified structure of the ontology; Candidate object attribute determination rules: If it expresses a stable, abstract, and reusable semantic relationship between two traffic knowledge objects, it is given priority as a candidate object attribute, and suggestions on the domain and range are given based on the simplified structure of the ontology. Candidate data attribute determination rules: If the expression is a static, stable, or configurable literal attribute that is attached to a certain type of object, it is given priority as a candidate data attribute, and suggestions on the domain and data type are given.

5. The traffic situation knowledge incremental evolution method based on ontology structure constraints as described in claim 1, characterized in that: In step S4, the large language model compares the normalized candidate elements with existing ontology elements, following semantic comparison rules including: Semantic equivalence judgment rule: Determine whether the candidate element and the existing ontology element express the same transportation concept, the same relationship, or the same attribute; Granularity consistency judgment rule: Determine whether the candidate element and the existing ontology element are at the same abstraction level and the same spatial scale; Coverage relationship judgment rule: Determine whether the existing ontology elements have fully expressed the core semantics of the candidate elements; Candidate elements are merged into existing ontology elements only if all three rules are met; otherwise, candidate elements are retained as elements to be updated.

6. The traffic situation knowledge incremental evolution method based on ontology structure constraints as described in claim 1, characterized in that: In step S5, the large language model is incrementally updated based on the complete ontology structure, following the following judgment and writing rules: Inclusion criteria: First, determine whether the element to be updated has independent modeling value. If its essence is a specific instance, algorithm process, model method, or research task, then output a decision to reject inclusion; if it has a stable semantic boundary and is related to the traffic situation business concept, then continue to perform structural positioning. Structural positioning rules include: Class element location rule: Select the parent class with the closest semantics and the most stable hierarchy from the complete ontology, and attach the class element to be updated to the parent class through the rdfs:subClassOf relationship; Object attribute location rules: Force the domain and value range of the object attribute to be updated to be anchored to the available class anchor point of the complete ontology, and establish the association through rdfs:domain and rdfs:range; Data attribute location rules: Force the domain of the data attribute to be updated to be anchored to the available class anchor point of the complete ontology, and bind its value type to a valid XSD data type through rdfs:range; Update decision and write rules: Based on the results of inclusion judgment and structural positioning, the final update decision is output; when the decision is that it is locatable, the elements to be updated and their structural positions are formalized into OWL semantic descriptions and written into the initial complete ontology; when the decision is questionable or rejected, no writing is performed, and the manual review interface is retained.

7. A traffic situation knowledge incremental evolution system based on ontology structure constraints, employing the traffic situation knowledge incremental evolution method based on ontology structure constraints as described in any one of claims 1 to 6, characterized in that, include: The corpus structuring and filtering module is configured to transcribe traffic documents into structured text, call a large language model to evaluate the chapter modeling value, and generate effective corpus units. The prior constraint extraction module is configured to extract a simplified ontology structure from the initial complete ontology, and under the prior constraints of the simplified ontology structure, call the large language model to extract structured candidate ontology elements from the effective corpus units. The two-level semantic alignment module is configured to perform internal repetitive convergence of candidate ontology elements by calculating vector similarity, and call the large language model to determine the equivalence, granularity and coverage relationship between candidate elements and existing ontology, and output the elements to be updated. The structure-aware evolution module is configured to call the large language model to determine the inclusion of elements to be updated and assign precise logical structure positions under the global constraints of the initial complete ontology structure. Finally, it writes these elements into the ontology through formal code to achieve evolutionary updates.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the traffic situation knowledge incremental evolution method based on ontology structure constraints as described in any one of claims 1 to 6.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the traffic situation knowledge incremental evolution method based on ontology structure constraints as described in any one of claims 1 to 6.

Images