Risk identification methods and electronic devices based on entity connection relationships

Abstract

This application provides a risk identification method and electronic device based on entity connection relationships, comprising: acquiring multi-source data of a mechanical system's physical interface and identifying target entity information and target entity connection relationship information in the multi-source data; performing semantic similarity matching between the target entity information and target entity connection relationship information and a failure rule base to obtain a semantic similarity score; generating a risk seed for the failure risk entity connection relationship when the semantic similarity score is greater than a preset threshold; and performing a failure risk assessment based on the risk seed to obtain the failure risk level of the failure risk entity connection relationship. This solves the problems of existing risk identification designs for entity connection relationships in mechanical system physical interfaces, such as single data analysis dimensions, low accuracy of physical interface risk identification, and complex reasoning mechanisms. It achieves automated identification of risks in mechanical system physical interfaces, improves the completeness of physical interface risk identification, and significantly enhances the ability to discover hidden risks.

Description

Technical Field

[0001] This application relates to the field of mechanical system risk identification technology, and in particular to a risk identification method and electronic device based on entity connection relationships. Background Technology

[0002] The existing risk identification design for physical interface entity connection relationships in mechanical systems suffers from problems such as single data analysis dimensions, low accuracy of physical interface risk identification, and complex reasoning mechanisms, making it difficult to meet the complex risk identification needs of mechanical systems. Summary of the Invention

[0003] This application provides a risk identification method and electronic device based on entity connection relationships, aiming to improve the existing design of risk identification for physical interface entity connection relationships in mechanical systems, which suffers from problems such as single data analysis dimension, low accuracy of physical interface risk identification, and complex reasoning mechanism.

[0004] This application proposes a risk identification method based on entity connection relationships, comprising: acquiring multi-source data of the physical interface of a mechanical system, and identifying target entity information and target entity connection relationship information in the multi-source data; performing semantic similarity matching between the target entity information and the target entity connection relationship information and a predefined failure rule base to obtain a semantic similarity score, and generating a risk seed for the failure risk entity connection relationship when the semantic similarity score is greater than a preset threshold; and performing a failure risk assessment based on the risk seed to obtain the failure risk level of the failure risk entity connection relationship.

[0005] Optionally, in some embodiments, before identifying target entity information and target entity connection relationship information in the multi-source data, the method includes: extracting at least one entity from the multi-source data; calculating the multi-feature similarity degree between the at least one entity, and determining whether there is a common entity among the at least one entity based on the multi-feature similarity degree between the at least one entity; if there is a common entity among the at least one entity, merging and deduplicating the common entity to obtain the target entity in the multi-source data.

[0006] Optionally, in some embodiments, calculating the multi-feature similarity between the at least one entity includes: comparing the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between the at least one entity; and calculating the multi-feature similarity between the at least one entity based on the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between the at least one entity.

[0007] Optionally, in some embodiments, determining whether there is a common entity among the at least one entities based on the multi-feature similarity between the at least one entity includes: identifying whether there is an entity among the at least one entities whose multi-feature similarity is greater than a preset similarity based on the multi-feature similarity between the at least one entity; if there is an entity among the at least one entities whose multi-feature similarity is greater than the preset similarity, then it is determined that there is a common entity among the at least one entities.

[0008] Optionally, in some embodiments, identifying the target entity information and target entity connection relationship information in the multi-source data includes: identifying the entity type and attributes of the target entities in the multi-source data, and identifying the connection relationship type and attributes of the physical connection relationship between the target entities.

[0009] Optionally, in some embodiments, after generating the risk seeds of the failure risk entity connection relationships, the method includes: semantically enhancing the risk seeds of the failure risk entity connection relationships using a preset vector library and a preset knowledge graph to generate a natural language risk description.

[0010] Optionally, in some embodiments, before semantically enhancing the risk seeds of the failure risk entity connection relationships using a preset vector library and a preset knowledge graph to generate a natural language risk description, the process includes: constructing a knowledge graph including entity-physical connection relationship-attribute based on the entity type and attributes of the target entities, the connection relationship type of the physical connection relationships between the target entities, and the attributes of the physical connection relationships; generating embedding vectors based on the text corpus in the multi-source data, and establishing an association index between the embedding vectors and the entities in the knowledge graph to obtain the vector library.

[0011] Optionally, in some embodiments, the step of obtaining the failure risk level of the failure risk entity connection relationship by performing failure risk assessment based on the risk seed includes: assessing the failure impact degree score of the failure risk entity connection relationship based on a first evaluation index, assessing the reliability score of the failure risk entity connection relationship based on a second evaluation index, and assessing the failure impact range score of the failure risk entity connection relationship based on a third evaluation index; calculating the failure risk value of the failure risk entity connection relationship based on the failure impact degree score, the reliability score, and the failure impact range score; and determining the failure risk level of the failure risk entity connection relationship based on the failure risk value of the failure risk entity connection relationship.

[0012] Optionally, in some embodiments, after obtaining the failure risk level of the failure risk entity connection relationship through failure risk assessment based on the risk seed, the process includes: generating a list of failure risk entity pairs according to the failure risk level of the failure risk entity connection relationship, generating a failure risk propagation path chain according to the failure risk entity connection relationship, and generating a connection risk analysis report according to the failure risk entity connection relationship; and pushing the list of failure risk entity pairs, the failure risk propagation path chain, and the connection risk analysis report to the target terminal.

[0013] This application also proposes a risk identification system based on entity connection relationships, comprising: an acquisition module for acquiring multi-source data of the physical interface of a mechanical system and identifying target entity information and target entity connection relationship information in the multi-source data; a risk identification module for performing semantic similarity matching between the target entity information and the target entity connection relationship information and a predefined failure rule base to obtain a semantic similarity score, and generating a risk seed for the failure risk entity connection relationship when the semantic similarity score is greater than a preset threshold; and a risk assessment module for performing failure risk assessment based on the risk seed to obtain the failure risk level of the failure risk entity connection relationship.

[0014] Optionally, in some embodiments, before identifying the target entity information and target entity connection relationship information in the multi-source data, the acquisition module is further configured to: extract at least one entity from the multi-source data; calculate the multi-feature similarity degree between the at least one entity, and determine whether there is a common entity among the at least one entity based on the multi-feature similarity degree between the at least one entity; if there is a common entity among the at least one entity, then merge and deduplicate the common entity to obtain the target entity in the multi-source data.

[0015] Optionally, in some embodiments, the acquisition module is further configured to: compare the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between the at least one entity; and calculate the multi-feature similarity degree between the at least one entity based on the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between the at least one entity.

[0016] Optionally, in some embodiments, the acquisition module is further configured to: identify whether there is an entity among the at least one entities whose multi-feature similarity is greater than a preset similarity based on the multi-feature similarity between the at least one entity; if there is an entity among the at least one entity whose multi-feature similarity is greater than the preset similarity, then determine that the same entity exists among the at least one entity.

[0017] Optionally, in some embodiments, the acquisition module is further configured to: identify the entity type and attributes of the target entities in the multi-source data, and identify the connection relationship type and attributes of the physical connection relationship between the target entities.

[0018] Optionally, in some embodiments, after generating risk seeds for the connection relationships of failure risk entities, the risk identification module is further configured to: semantically enhance the risk seeds for the connection relationships of failure risk entities using a preset vector library and a preset knowledge graph to generate a natural language risk description.

[0019] Optionally, in some embodiments, before semantically enhancing the risk seeds of the failure risk entity connection relationships using a preset vector library and a preset knowledge graph to generate a natural language risk description, the risk identification module is further configured to: construct a knowledge graph including entity-physical connection relationship-attribute based on the entity type and attributes of the target entity, the connection relationship type of the physical connection relationship between the target entities, and the attributes of the physical connection relationship; generate embedding vectors based on the text corpus in the multi-source data, and establish an association index between the embedding vectors and the entities in the knowledge graph to obtain the vector library.

[0020] Optionally, in some embodiments, the risk assessment module is further configured to: assess the failure impact score of the failure risk entity connection relationship based on a first evaluation index, assess the reliability score of the failure risk entity connection relationship based on a second evaluation index, and assess the failure impact range score of the failure risk entity connection relationship based on a third evaluation index; calculate the failure risk value of the failure risk entity connection relationship based on the failure impact score, the reliability score, and the failure impact range score; and determine the failure risk level of the failure risk entity connection relationship based on the failure risk value of the failure risk entity connection relationship.

[0021] Optionally, in some embodiments, after obtaining the failure risk level of the failure risk entity connection relationship through failure risk assessment based on the risk seed, the risk assessment module is further configured to: generate a list of failure risk entity pairs according to the failure risk level of the failure risk entity connection relationship, generate a failure risk propagation path chain according to the failure risk entity connection relationship, and generate a connection risk analysis report according to the failure risk entity connection relationship; and push the list of failure risk entity pairs, the failure risk propagation path chain, and the connection risk analysis report to the target terminal.

[0022] This application also proposes an electronic device, including a processor and a memory, wherein the memory is used to store computer programs; and the processor is used to execute the programs stored in the memory to implement the risk identification method based on entity connection relationships.

[0023] This application also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the risk identification method based on entity connection relationships described in any embodiment of this application. Attached Figure Description

[0024] Figure 1 This is a flowchart of a risk identification method based on entity connection relationship provided in an embodiment of this application; Figure 2 This is a schematic diagram of the system architecture involved in the risk identification method based on entity connection relationship provided in an embodiment of this application; Figure 3 This is a schematic diagram of the reasoning process of the analysis and reasoning module provided in an embodiment of this application; Figure 4 This is a structural diagram of the risk identification system based on entity connection relationships provided in the embodiments of this application; Figure 5 This is a structural diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0025] To make the technical problems, technical solutions, and beneficial effects solved by this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0026] Currently, some patents combine AI and knowledge graphs for risk identification, such as the patent: Safety Production Risk Identification Method and System Based on Knowledge Graph. This patent aims to generate embedded representations by aggregating neighborhood information of entity nodes and relation edges in a knowledge graph through graph neural networks, calculate node weights using attention mechanisms to identify risk patterns, traverse associated paths to generate risk propagation sequences, dynamically update entity states by fusing temporal features, and finally complete risk identification by iteratively querying and integrating environmental factors.

[0027] However, this solution has the following technical problems: 1) The risk identification targets are too generalized and lack specificity: The solution focuses on multiple entities such as equipment, personnel and environment at the same time, which leads to the dispersion of risk identification targets. It is not accurate enough in identifying risks related to the connection relationships of mechanical system entities, and it is difficult to deeply explore the risks related to the connection relationships.

[0028] 2) Single data source input: This solution directly obtains entity and relationship data from the preset knowledge graph database interface. Its analysis effect is heavily dependent on the completeness and accuracy of the upstream knowledge graph construction. If the knowledge graph is missing or incorrect, it will directly affect the risk identification results.

[0029] 3) Complex reasoning mechanism, posing performance and interpretability challenges: The core reasoning of this scheme heavily relies on complex deep learning models such as graph neural networks (GNNs) and attention mechanisms. These models have high computational overhead and high response latency, hindering real-time analysis. Furthermore, its decision-making process is like a "black box," exhibiting poor interpretability and low confidence, making it difficult to provide sufficient evidence and clear guidance for design improvements.

[0030] Therefore, existing designs for risk identification based on physical interface entity connection relationships in mechanical systems suffer from problems such as limited data analysis dimensions, low accuracy of physical interface risk identification, and complex reasoning mechanisms. This application provides a risk identification method based on entity connection relationships.

[0031] It should be noted that the system architecture involved in the embodiments of this application is as follows: Figure 2 As shown, it includes an input module, a data processing and fusion module, an analysis and reasoning module, and an output and early warning module.

[0032] 1) Input Module: This module takes diverse and heterogeneous data, including documents, 2D drawings, and 3D models, as input to achieve the parsing and fusion of multimodal data. It avoids the limitations of traditional methods that rely on a single data source or human experience. The input module focuses on data cleaning, parsing, and output functions, providing standardized input for subsequent modules.

[0033] 2) Data Processing and Fusion Module: Receives standardized data from the input module and constructs a knowledge graph and vector library for risk analysis through a serialized processing flow.

[0034] 3) Analysis and Reasoning Module: Based on the entity connection relationship of the knowledge graph and the semantic context of the vector library, it adopts a collaborative reasoning architecture led by the rule engine and assisted by the large language model to identify and assess direct failure risk and indirect failure risk.

[0035] 4) Output and Early Warning Module: By integrating visualization and correlation system interfaces, the structured output of the analysis and reasoning module is transformed into multi-level, actionable risk decision support information, thereby realizing closed-loop management from risk perception to response.

[0036] Specifically, risk identification methods based on entity connection relationships include... Figure 1 As shown, it includes the following steps: In step S101, multi-source data of the physical interface of the mechanical system is acquired, and target entity information and target entity connection relationship information in the multi-source data are identified.

[0037] The multi-source data of the physical interface of the mechanical system includes diverse and heterogeneous data such as documents, two-dimensional drawings, and three-dimensional digital models.

[0038] Among them, the 3D digital model is responsible for providing accurate geometric assembly relationships, interference checks, and quality attribute analysis.

[0039] 1) Input file types: including STEP (Standard for the Exchange of Product modeldata), IGES (Initial Graphics Exchange Specification), or native CAD format, etc.

[0040] 2) Parsing Technology: For exchange data formats such as STEP, IGES, and JT, software such as the 3D geometric modeling kernel OPENCASCADE is used to read the geometric data and topological information in the file, and then parse and convert it according to the standard data structure, transforming it into an internal data structure that the program can process. For native CAD formats, such as .CATPart, .CATProduct, .sldprt, .sldasm, .prt, .asm, and .par, it is necessary to call their native CAD software API for parsing, and obtain the geometric information and assembly relationships of the model through the interface functions provided by the API.

[0041] 3) Output: Uniform JSON structure data, including: Assembly tree: Uses a hierarchical structure to clearly define the parent-child and sibling relationships between entities; Geometric constraint pair list: Contains unique identifiers of two faces / edges / points that have geometric constraints (such as fit, alignment, coaxial) and their constraint types; Attribute list: includes part number, part name, material, mass, density, and description.

[0042] 2. Two-dimensional drawings: responsible for providing information that may not be directly included in the three-dimensional model, such as dimensional tolerances, geometric tolerances, process notes, and key technical requirements.

[0043] 1) File types include native vector formats (.dwg, .dxf, .drw, .frm, .idw, .dwg, .CATDrawing, .slddrw, .prt, .asm, .ipt, etc.), general formats (.pdf, .step, .iges, .svt, etc.), and image formats (.tiff, .png, .jpeg, etc.). 2) Analysis Technology: Computer vision (CV) and optical character recognition (OCR) technologies are used to identify graphics, symbols, and text in drawings. For graphics and symbols, a trained image recognition model is used for recognition; for text, OCR technology is used to extract and understand its semantics (such as tolerance symbols, roughness symbols, and welding symbols), and a symbol semantic database is established for matching and understanding.

[0044] 3) Output: JSON structured data, including: The associated entity ID; tolerance type and specific value; process annotation text content; assembly technical requirements.

[0045] 3. Documentation: Responsible for providing design documents, BOM (Bill of Materials), interface control documents, design approvals, design specifications, FMEA (Failure Mode and Effects Analysis), maintenance manuals, product manuals, test reports, etc. In other words, the document input should filter out content that is not related to the entity and only extract design documents, FMEAs, etc. that involve the connection relationship between entities, so as to keep the risk identification focused and avoid noise interference.

[0046] 1) Input file types: docx, json, excel, pdf, jpg, txt, markdown, html, etc.

[0047] 2) Parsing technology: Different parsing libraries are used to parse different file formats; text extraction uses named entity recognition technology in natural language processing to extract key entities (such as part numbers and tolerance requirements); block segmentation is based on semantic similarity algorithms (such as Sentence-BERT) and paragraph structure rules to ensure the semantic integrity of text blocks.

[0048] 3) Output: A collection of text blocks, including information such as text paragraphs, document source, and chapter position.

[0049] The input module ultimately outputs three types of standardized JSON data: 1) Geometric and structural data: Integrates all assembly levels and geometric constraint pairs for building the topological skeleton of the knowledge graph.

[0050] 2) Symbol and text data: All identified drawing symbols and text annotations are integrated as the basis for entity attribute and relationship classification.

[0051] 3) Text corpus: The collection of text files after being segmented and their metadata, used to build a vector library and assist in the construction of a knowledge graph.

[0052] Optionally, in some embodiments, before identifying target entity information and target entity connection relationship information in multi-source data, the method includes: extracting at least one entity from the multi-source data; calculating the multi-feature similarity degree between at least one entity, and determining whether there is a common entity among at least one entity based on the multi-feature similarity degree between at least one entity; if there is a common entity among at least one entity, merging and deduplicating the common entity to obtain the target entity in the multi-source data.

[0053] To resolve the discrepancies in the identification of the same physical entity across multiple data sources and to improve its attributes, entity alignment and attribute fusion are required. Specifically, this involves identifying whether the same physical entity exists in the multiple data sources. If the same physical entity exists, it needs to be merged and deduplicated to form a complete and unified set of attributes for that entity.

[0054] Specifically, a four-dimensional feature fusion decision method is adopted, which compares the identifier matching degree, geometric feature matching degree, context matching degree and functional matching degree between at least one entity to calculate the multi-feature similarity degree between at least one entity, and determines whether there is the same physical entity among at least one entity based on the multi-feature similarity degree between at least one entity.

[0055] By employing the aforementioned technical solutions, the existence of the same entity within at least one entity is determined, avoiding entity confusion and omissions caused by differences in data sources, thus improving the accuracy and completeness of entity identification. Multiple records identified as the same physical entity are merged and deduplicated to form a complete, unified, and standardized set of entity attributes, providing high-quality, highly reliable foundational data for subsequent risk analysis. Through the merging, deduplication, and completion of globally unique entity identifiers and attributes, duplicate nodes and redundant relationships are prevented from interfering with subsequent connection relationship identification and risk reasoning.

[0056] Optionally, in some embodiments, calculating the multi-feature similarity between at least one entity includes: comparing the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between at least one entity; and calculating the multi-feature similarity between at least one entity based on the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between at least one entity.

[0057] Specifically, identifier matching, geometric fingerprint matching, topological context matching, and functional matching are performed between entities to obtain at least one identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between entities. The specific matching methods are as follows: (1) Identifier Matching Degree (IMD): This involves comparing unique identifiers such as part numbers and model numbers. If they are completely identical, the highest weight is assigned. This is achieved through a string exact matching algorithm.

[0058] (2) Geometric Matching Degree (GMD) is the calculation of the similarity of geometric features (such as volume, key dimensions, and centroid position) of entities. A higher matching degree is assigned a higher weight. Geometric similarity calculation algorithms, such as Euclidean distance, are used.

[0059] (3) Topological Matching Degree (TMD): This involves comparing the position of entities in the assembly tree and their connections. Entities with similar topological environments are assigned higher weights. Topological context similarity is determined by analyzing the hierarchical structure and connections of the assembly tree.

[0060] (4) Functional Matching Degree (FMD): Based on natural language processing technology, it compares the semantic similarity of entity function description texts from documents and drawings. It adopts semantic similarity calculation models in natural language processing, such as cosine similarity.

[0061] Based on the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree mentioned above, a multi-feature fusion similarity (MFFS) score is generated by combining the similarity scores of each item. MFFS = f(IMD, GMD, TMD, FMD), where the function f determines the weight allocation based on the actual situation, such as using a weighted average method.

[0062] By employing the aforementioned technical solution, and through the comprehensive similarity analysis of four dimensions—identifier matching, geometric fingerprint matching, topological context matching, and functional matching—a multi-dimensional feature-based comprehensive judgment of physical entities is achieved. This avoids the problem of single feature matching being susceptible to differences in data sources, effectively reduces entity confusion, duplicate identification, and omissions, and improves the accuracy and robustness of entity alignment.

[0063] Optionally, in some embodiments, determining whether there is a common entity among at least one entity based on the multi-feature similarity between at least one entity includes: identifying whether there is an entity among at least one entity whose multi-feature similarity is greater than a preset similarity based on the multi-feature similarity between at least one entity; if there is an entity among at least one entity whose multi-feature similarity is greater than the preset similarity, then it is determined that there is a common entity among at least one entity.

[0064] The preset similarity can be a threshold set by the user, a threshold obtained through a limited number of experiments, or a threshold obtained through a limited number of computer simulations; no specific limitation is made here.

[0065] When the Multi-Feature Similarity Score (MFFS) exceeds the preset similarity score, the system automatically determines that the entity records point to the same physical entity and assigns it a globally unique entity identifier.

[0066] When the Multi-Feature Similarity (MFFS) falls below a certain threshold or a rule conflict occurs, the system automatically marks the relevant entity records and submits them to domain experts for final adjudication.

[0067] The attributes of multiple records from different data sources that are identified as the same entity are merged and deduplicated to form a complete and unified set of attributes for the entity, thereby obtaining the target entity in the multi-source data. This allows for further identification of entity information and entity connection relationship information of the target entity in the multi-source data.

[0068] For example, the system matches "shoulder" entities appearing in multi-source data: Identifier matching: The part numbers in the 3D model, 2D drawings, and BOM are completely consistent, which scores the highest.

[0069] Geometric fingerprint matching: volume, outer diameter, length, and centroid position are consistent.

[0070] Topology context matching: The assembly relationship is "bushel - control arm - subframe".

[0071] Function matching: The document descriptions are all "suspension motion compensation, vibration damping".

[0072] Furthermore, multi-feature similarity is calculated: MFFS=f(IMD,GMD,TMD,FMD); If the result is greater than the preset similarity, the system determines that they are the same physical entity.

[0073] By using the above technical solution, and by using a preset similarity threshold as the criterion, entities with multiple feature similarities exceeding the threshold are automatically identified and determined to be the same physical entity. This significantly improves the automation efficiency of entity alignment and reduces the cost of manual intervention. At the same time, for scenarios where the MFFS is below the threshold or there are rule conflicts, a domain expert adjudication mechanism is introduced to avoid misjudgment and omission, ensuring the accuracy of the determination of the same entity and adapting to complex scenarios with multi-source heterogeneous data.

[0074] Optionally, in some embodiments, identifying target entity information and target entity connection relationship information in multi-source data includes: identifying the entity type and attributes of the target entities in the multi-source data, and identifying the connection relationship type and attributes of the physical connection relationships between the target entities.

[0075] The specific steps for identifying target entity information and target entity connection relationship information in multi-source data are as follows: First, structured information is extracted from multi-source data based on the schema. The structured information mainly consists of: (1) Entity type: refers to physical objects with specific shape, material, position and engineering function in the geometric space of the whole vehicle, such as parts, subsystems, assemblies, etc.; (2) Physical connection relationship type: Focusing on the physical connection relationship between entities, it is defined as three major categories: structural fixed type (rigid connection used to transmit the main load, such as welding and bolts), motion compensation type (connection that allows relative motion or provides flexible compensation, such as hinges and bushings), and medium transmission type (connection used to transmit fluid, energy or signal, such as pipes and wire harnesses).

[0076] (3) Attribute set: Defines the attribute fields of entities and relations. The attributes of entities include part number, name, material, mass, density, description text, etc.; the attributes of relations include connection type, tolerance, preload, etc., providing detailed information for entities and relations.

[0077] Furthermore, entity and relation extraction is performed based on the schema, including: (1) Entities and Attributes: Identify and extract entities that conform to the schema definition from the assembly tree, part numbers, drawing annotations and document descriptions, and assign them preliminary types and attributes. Entity types and attributes are determined through techniques such as keyword matching and semantic analysis.

[0078] (2) Physical connection relationship and attributes: Based on the geometric constraint pairs (such as surface mating, coaxial) parsed by the input module and the co-occurrence description in the document, the physical or functional association between entities is identified as the unclassified original relationship, and its related attributes (such as constraint type) are recorded.

[0079] Through the above technical solution, entities and their attributes, physical connection relationships and their attributes are automatically extracted from multi-source data based on schema. This enables the comprehensive extraction of physical entities such as vehicle parts, subsystems, and assemblies, as well as three major categories of physical connection relationships: structural fixed type, motion compensation type, and medium transmission type. This provides complete and accurate basic elements for subsequent knowledge graph construction and physical interface risk identification.

[0080] In step S102, semantic similarity matching is performed between the target entity information and the target entity connection relationship information and the predefined failure rule base to obtain a semantic similarity score. When the semantic similarity score is greater than a preset threshold, a risk seed for the failure risk entity connection relationship is generated.

[0081] Specifically, after identifying the target entity information and the target entity connection relationship information, the physical connection relationships are classified to support subsequent risk analysis.

[0082] Specifically, the system first identifies the physical connections between entities based on the geometric constraint pairs (such as face matching and coaxiality) provided by the input module, combined with the association information extracted from the document. The specific implementation steps include: (1) Establish a list of key components: Based on FMEA, design inputs and historical failure data, predefine a batch of key components as “seeds”; (2) Radial scanning: In the 3D digital model and knowledge graph, take each key component as the central node and scan outwards all entities of its first-degree connection (direct connection) and second-degree connection (neighbor's neighbor); the scanning operation is implemented by graph traversal algorithm.

[0083] (3) Simultaneously focus on identifying intra-system and inter-system connections. By analyzing the systems to which entities belong and their connection relationships, determine whether the connection is intra-system or inter-system, so as to construct a risk identification network with distinct levels from local to global.

[0084] Further classification of physical connection relationships: According to the predefined schema classification, based on the predefined classification rules and their geometric and functional characteristics, they are classified into specific physical connection relationship types (such as structural fixed type, motion compensation type, and medium transmission type).

[0085] Optionally, in some embodiments, after generating the risk seeds of the failure risk entity connection relationships, the method includes: semantically enhancing the risk seeds of the failure risk entity connection relationships using a preset vector library and a preset knowledge graph to generate a natural language risk description.

[0086] Optionally, in some embodiments, before semantically enhancing the risk seeds of the connection relationships between failure risk entities using a preset vector library and a preset knowledge graph to generate a natural language risk description, the process includes: constructing a knowledge graph including entity-physical connection relationship-attribute based on the entity type and attributes of the target entity, the connection relationship type of the physical connection relationship between the target entities, and the attributes of the physical connection relationship; generating embedding vectors based on text corpus from multi-source data, and establishing an association index between the embedding vectors and entities in the knowledge graph to obtain the vector library.

[0087] The specific method for constructing a knowledge graph is as follows: using aligned and merged entities as nodes, and classified physical connections as edges, all attributes are integrated to construct a knowledge graph that includes entities, physical connections, and attributes.

[0088] The vector library is constructed as follows: the text corpus provided by the input module is preprocessed, the text is segmented according to semantics and embedded vectors are generated, and an association index between text fragments and entities in the knowledge graph is established to provide semantic support for the analysis and reasoning module.

[0089] Furthermore, this application adopts a collaborative reasoning architecture led by a rule engine and assisted by a large language model. The rule engine performs pattern matching and traversal on the knowledge graph, triggers predefined geometric, physical and empirical failure rules, and generates risk seeds for the connection relationship of failure risk entities.

[0090] The specific steps for generating risk seeds for the connection relationships of failure risk entities are as follows: Figure 3 As shown: Step 1: The rule engine is responsible for pattern matching and traversal on the knowledge graph, triggering predefined failure rules, and performing deterministic, logic- and algorithm-based reasoning.

[0091] Step 2: Based on a predefined failure rule base (including geometric, physical, and empirical rules), perform semantic similarity matching on entities and physical connections in the knowledge graph. To handle the diversity of engineering terminology, semantic similarity calculation based on word vectors or sentence vectors is used to achieve semantic-level fuzzy matching. In particular, ontology semantic normalization can be achieved by establishing an engineering terminology dictionary; Step 3: Risk Seed Generation: When the semantic similarity score exceeds a preset threshold, the corresponding failure rule is triggered and a structured "risk seed" is generated. The risk seed specifically includes: (1) Risk ID: Unique identifier.

[0092] (2) Risk Entity Sequence: An array of entity IDs, listed in the order of risk propagation. For directly failed entities, there is a pair of entity IDs, the sequence length is 2, and the corresponding edge in the knowledge graph is 1; for indirectly failed entities, all key entity IDs are listed in the order of propagation path, the sequence length is >2, and the corresponding inference chain has a path length >1.

[0093] (3) Triggering rule ID: The identifier of the specific invalidation rule that is triggered. This identifier can clearly point to a geometric, physical or empirical rule in the rule base.

[0094] (4) Preliminary evidence: a description of quantitative or qualitative evidence based on data calculations.

[0095] It should be noted that the predefined failure rule base is a set of structured logical rules defined by domain experts, used to systematically identify potential failures caused by entity connection relationships. The large model automatically discovers potential failure modes and rules through semantic analysis of entities and relationships in the knowledge graph, as well as clustering and classification of semantic vectors in the vector library. Combining historical data and expert feedback, the rules in the rule base are continuously extracted and dynamically optimized.

[0096] The failure rules in the predefined failure rule base can be specifically divided into: 1. Geometry-based failure rules: Risk assessment is based on the shape, size, and spatial relationships of the entity, specifically including: 1) Interference and Clearance Rules: This covers static interference checks to identify overlaps between entities in an assembled state; dynamic clearance checks to verify minimum clearance requirements for moving parts; and tolerance analysis to assess the impact of cumulative tolerances in the dimensional chain on fit performance.

[0097] 2) Envelope and layout rules: This includes envelope compliance checks to ensure that entities do not exceed the design envelope space; and accessibility checks to verify the operability of maintenance tools on critical components.

[0098] 3) Geometric feature rules: Identify geometric features that are prone to stress concentration, such as acute angles and grooves; and analyze the impact of surface roughness on key properties such as sealing performance.

[0099] 2. Physics-based failure rules: These rules determine failures based on physical principles such as mechanics, thermodynamics, fluid dynamics, and electromagnetism. Specifically, they include: 1) Structural interface: This includes strength failure assessment, checking whether the stress exceeds the material yield limit; stiffness failure assessment, evaluating whether the deformation affects the functional realization; and stability failure assessment, identifying buckling or resonance risks.

[0100] 2) Thermal interface: including thermal expansion mismatch analysis to assess the deformation compatibility of different materials under temperature changes; and overheat failure judgment to check whether the component temperature exceeds the material's tolerance limit.

[0101] 3) Fluid interface: This includes pressure loss analysis to determine whether the flow channel design leads to insufficient flow; and identification of cavitation and erosion risks.

[0102] 4) Electromagnetic interface: including interference failure judgment, analysis of electromagnetic compatibility issues; and insulation failure inspection, assessing whether the electric field strength exceeds the insulation material's tolerance.

[0103] 3. Additional judgments based on experience-based failure rules. These rules originate from accumulated practical knowledge and normative documents in the domain, specifically including: 1) Historical FMEA; 2) Regulations and industry standards; 3) Enterprise standards, design guidelines, and design manuals; 4) Product requirement specifications, etc.

[0104] For example, the rule engine performs rule matching on the knowledge graph: Matching according to geometric rules: check dynamic gaps and cumulative dimensional chain tolerances to determine the risk of insufficient interference.

[0105] Matching according to physical rules: Structural interface rule determination: There is insufficient strength and fatigue failure risk at the bonding surface between the bushing rubber and the metal; Thermal interface rule determination: Temperature changes may lead to thermal expansion mismatch.

[0106] Match according to empirical rules: match historical FMEA and design specifications to confirm that the interface is a high-priority interface.

[0107] When the semantic similarity score exceeds a threshold, the system generates a risk seed for the entity connection relationship of failure risk, which includes: Risk ID; Risk entity sequence, such as: [Bushing ID, Control Arm ID]; Trigger rule ID, such as: structural interface failure rule, hot interface failure rule; Preliminary evidence includes: insufficient over-exchange, insufficient dynamic gaps, and high-risk items in historical FMEA.

[0108] By constructing a knowledge graph with integrated entities as nodes, classified physical connection relationships as edges, and containing complete attributes, and combining it with a semantic vector library, the direct assembly relationship and indirect functional coupling relationship between entities can be fully represented, achieving complete coverage of explicit interface risks and implicit association risks, and fundamentally solving the problem of blind spots in traditional methods.

[0109] Optionally, in some embodiments, after generating the risk seeds of the failure risk entity connection relationships, the method includes: semantically enhancing the risk seeds of the failure risk entity connection relationships using a preset vector library and a preset knowledge graph to generate a natural language risk description.

[0110] Specifically, after generating risk seeds that establish the connections between entities at risk of failure, the large language model uses natural language processing (NLP) techniques to semantically understand and expand these risk seeds generated by the rule engine. Combining the topological relationships of the knowledge graph with the semantic context of the vector library, the structured risk seeds are transformed into complete natural language risk descriptions containing reasoning chains. By analyzing the potential associations between entities and relationships in the knowledge graph, the large model may discover overlooked potential risks. These risks will serve as reference information for experts to optimize the rule base and will not be directly incorporated into the final risk decision.

[0111] Through the above technical solutions, the large language model transforms risk seeds containing structured information such as IDs and risk entity sequences into complete natural language risk descriptions that include reasoning chains through semantic understanding and expansion. This clearly presents the entities involved in the risk, their connections, triggering causes, and potential impacts, enabling technical personnel and decision-makers to quickly understand risk information.

[0112] In step S103, failure risk assessment is performed based on risk seeds to obtain the failure risk level of the connection relationship between failure risk entities.

[0113] Optionally, in some embodiments, the failure risk level of the failure risk entity connection relationship is obtained by performing a failure risk assessment based on the risk seed, including: assessing the failure impact degree score of the failure risk entity connection relationship based on a first evaluation index, assessing the reliability score of the failure risk entity connection relationship based on a second evaluation index, and assessing the failure impact range score of the failure risk entity connection relationship based on a third evaluation index; calculating the failure risk value of the failure risk entity connection relationship based on the failure impact degree score, reliability score, and failure impact range score; and determining the failure risk level of the failure risk entity connection relationship based on the failure risk value of the failure risk entity connection relationship.

[0114] Among them, the first evaluation indicator can be a critical evaluation indicator, the second evaluation indicator is a robustness evaluation indicator, and the third evaluation indicator is a broad evaluation indicator.

[0115] Specifically, the failure risk value adopts a three-dimensional quantitative model and follows the following principles: 1) Targeted principle: Only failure risks directly related to the connection relationship of the entity are evaluated, and non-connection dimensions such as material defects and manufacturing processes are not covered; 2) Relativity principle: Risk priority is calculated by scoring the three dimensions of criticality, robustness and wide impact of the connection.

[0116] Critical performance indicators (CPIs) are used to assess the impact of connection failures on the functional integrity and security of the system. The score for the degree of failure impact ranges from 1 to 10, with higher scores indicating a more critical impact of the connection on the normal operation and security of the system.

[0117] Robustness evaluation metrics are used to assess the ability of a connection structure to resist potential failure factors such as design deviations, wear, and vibration. Reliability scores range from 1 to 10, with higher scores indicating higher reliability and less susceptibility to failure.

[0118] The pervasiveness evaluation index is used to assess the potential scope, speed, and difficulty of recovery once a connectivity failure occurs, as it propagates along the connectivity network. The failure impact score ranges from 1 to 10 points, with higher scores indicating a wider potential impact and more severe consequences.

[0119] Furthermore, based on the failure impact score, reliability score, and failure impact range score of the assessed failure risk entity connection relationship, the failure risk value of the entity connection relationship is calculated. Specifically, the failure risk value is calculated as follows: (Failure impact score × Failure impact range score) / Reliability score.

[0120] The severity is calculated by multiplying the failure impact score by the failure impact range score, and 1 / reliability score represents the probability. The calculation results are categorized into high-risk, medium-risk, and low-risk levels on a scale of 1-100. This model focuses on the specific risk source of connectivity, and its assessment scope serves as a necessary supplement and refinement to comprehensive analysis methods (such as DFMEA), thereby providing a quantitative basis for design improvement and test resource allocation.

[0121] The output of the analysis and reasoning module provides a structured data foundation for subsequent early warning and response. Its output includes: (1) Risk seed: Core risk data encapsulated in JSON format, including risk ID, risk entity sequence, trigger rule ID and preliminary evidence.

[0122] (2) Complete risk language description: a natural language risk description that includes reasoning chain and is generated by a large model based on risk seeds.

[0123] (3) Risk score and level: The quantitative risk value and the corresponding failure risk level (high risk level, medium risk level, low risk level) calculated based on the three-dimensional scoring model.

[0124] (4) High-risk entities and connection list: a structured list of key entity pairs and their connection relationships sorted by risk score.

[0125] For example, let's rate a physical connection interface: Key performance indicators: 9 points; Robustness performance indicators: 4 points; Broadness performance indicators: 7 points.

[0126] The failure risk value is (9×7) / 4=15.75, and the system determines it to be a high-risk level.

[0127] Furthermore, the final output of the analysis and reasoning module is: (1) Structured risk seeds; (2) Natural language risk description; (3) Risk value 15.75, high risk level; (4) List of high-risk entities: bushing-control arm.

[0128] The above technical solution uses the quantitative formula Risk Value = (Failure Impact Score × Failure Impact Range Score) / Reliability Score to transform abstract risks into calculable and comparable specific values. At the same time, risk levels are divided according to a scale of 1-100, making the risk assessment results more objective and scientific, and avoiding the subjective bias of manual assessment.

[0129] Optionally, in some embodiments, after obtaining the failure risk level of the connection relationship of failure risk entities based on the failure risk assessment based on the risk seed, the process includes: generating a list of failure risk entity pairs according to the failure risk level of the connection relationship of failure risk entities, generating a failure risk propagation path chain according to the connection relationship of failure risk entities, and generating a connection risk analysis report according to the connection relationship of failure risk entities; and pushing the list of failure risk entity pairs, the failure risk propagation path chain, and the connection risk analysis report to the target terminal.

[0130] Specifically, after assessing failure risk based on risk seeds to obtain the failure risk level of the entity connection relationship, the output and early warning module, through integrated visualization and association system interfaces, transforms the structured output of the analysis and reasoning module into multi-level, actionable risk decision support information. This output and early warning module includes a dashboard with functions such as listing high-risk items and visualizing risk distribution. Regarding risk item listing, the dashboard can accurately and comprehensively list the high-risk items identified by the system based on multiple filtering conditions such as risk level, type, and occurrence time. It also provides detailed information for each item, including risk description, potential impact range, and estimated potential losses, offering decision-makers a clear risk list. For risk distribution visualization, the dashboard uses intuitive chart formats such as map heatmaps, bar charts, and pie charts to clearly present the risk distribution trends in different regions, business processes, and time periods, helping users quickly locate concentrated risk areas and key processes for targeted risk control.

[0131] Specifically, it includes: 1. Dynamic Risk Presentation: Provides macro-level situational awareness for overall risk monitoring, including: 1) List of high-priority risk entities; 2) Key failure propagation chain and evidence chain diagrams, namely, a visual graph of risk propagation path generated based on knowledge graph. Nodes represent entities, and edges are labeled with connection relationships and risk types. It supports interactive expansion of reasoning chains, such as bushing failure → control arm offset → wheel alignment deviation → driving instability. 3) Visualized global risk report integrating risk and reasoning chain.

[0132] 2. Provide in-depth analysis to drive actual business processes, including: 1) A comprehensive analysis report on vehicle-level connection risks is presented, which categorizes connections into structurally fixed, motion-compensated, and media-transmitted types. The most vulnerable connection interfaces and the highest-risk propagation paths are highlighted to provide data support for design improvements and testing priorities. 2) Connect with the enterprise's quality management system and project management system to automatically push risk warnings and reports, including fields such as risk ID, entity sequence, and handling suggestions, to relevant quality managers and design engineers, triggering design prevention or rectification measures.

[0133] The above technical solutions generate a list of high-risk entity pairs, a chain of failure risk propagation paths, and connect risk analysis reports based on risk levels. This helps users quickly locate areas of concentrated risk, key weak interfaces, and core propagation paths, significantly improving risk monitoring efficiency. By connecting to enterprise quality management, project management, and other systems through interfaces, risk warnings, risk lists, and analysis reports are automatically pushed to target terminals and responsible personnel, forming a closed-loop management system, shortening the risk response cycle, and improving the efficiency of rectification and implementation.

[0134] In summary, the beneficial effects of this application are as follows: (1) Based on the combined effect of multi-source data input, collaborative construction of knowledge graph and vector library, and dual-drive reasoning mechanism, the system achieves complete coverage of physical interface risk identification. The input module ensures comprehensive collection of physical interfaces through multi-source data fusion; the data processing module complements the structured representation of knowledge graph and the semantic description of vector library, fully describing direct assembly relationships and indirect functional coupling relationships; the analysis and reasoning module achieves systematic identification of explicit interface risks and implicit association risks through the collaboration of deterministic reasoning of rule engine and semantic understanding of large model. This multi-level collaborative mechanism fundamentally solves the identification blind spot problem that is difficult to avoid in traditional manual review and single-method analysis.

[0135] (2) Based on the dual-drive architecture of rule engine and big model, rule engine ensures strict execution of explicit design rules and geometric constraints, while big model discovers potential indirect related risks through understanding semantic context. While maintaining the traceability of deterministic reasoning, the system realizes in-depth mining of complex risk patterns, forming an organic unity of reliability assurance and discovery depth expansion.

[0136] (3) By establishing a closed-loop management process from risk identification to rectification verification, risk warnings based on connection relationship classification can be accurately pushed to relevant responsible personnel, realize automatic tracking of rectification progress, greatly shorten the cycle from risk identification to implementation of engineering measures, and significantly improve risk management efficiency.

[0137] (4) By transforming the expert experience in the field of physical interfaces into a structured rule base and continuously optimizing it based on the large model suggestions in engineering practice, a dynamic accumulation system for risk knowledge of connection relationships has been systematically constructed, laying a solid foundation for the continuous improvement of enterprise risk management capabilities.

[0138] According to the risk identification method based on entity connection relationships provided in this application, multi-source data of the physical interface of a mechanical system is acquired, and target entity information and target entity connection relationship information in the multi-source data are identified. Semantic similarity matching is performed between the target entity information and target entity connection relationship information and a predefined failure rule base to obtain a semantic similarity score. When the semantic similarity score is greater than a preset threshold, a risk seed for the failure risk entity connection relationship is generated. A failure risk assessment is performed based on the risk seed to obtain the failure risk level of the failure risk entity connection relationship. This solves the problems of single data analysis dimensions, low accuracy of physical interface risk identification, and complex reasoning mechanisms in existing risk identification designs for physical interface entities in mechanical systems. It achieves automated and systematic identification of risks in the physical interface of mechanical systems, improves the completeness of physical interface risk identification, and significantly enhances the ability to discover hidden risks.

[0139] This application also provides a risk identification system based on entity connection relationships. Please refer to... Figure 4 The risk identification system 10 based on entity connection relationships includes: an acquisition module 100, a risk identification module 200, and a risk assessment module 300.

[0140] The acquisition module 100 is used to acquire multi-source data of the physical interface of the mechanical system and identify target entity information and target entity connection relationship information in the multi-source data; the risk identification module 200 is used to perform semantic similarity matching between the target entity information and target entity connection relationship information and a predefined failure rule base to obtain a semantic similarity score, and generate a risk seed for the failure risk entity connection relationship when the semantic similarity score is greater than a preset threshold; the risk assessment module 300 is used to perform failure risk assessment based on the risk seed to obtain the failure risk level of the failure risk entity connection relationship.

[0141] Optionally, in some embodiments, before identifying target entity information and target entity connection relationship information in multi-source data, the acquisition module 100 is further configured to: extract at least one entity from the multi-source data; calculate the multi-feature similarity degree between at least one entity, and determine whether there is a common entity among at least one entity based on the multi-feature similarity degree between at least one entity; if there is a common entity among at least one entity, then merge and deduplicate the common entity to obtain the target entity in the multi-source data.

[0142] Optionally, in some embodiments, the multi-feature similarity calculation module 100 is further configured to: compare the identifier matching degree, geometric feature matching degree, context matching degree and functional matching degree between the at least one entity; and calculate the multi-feature similarity degree between the at least one entity based on the identifier matching degree, geometric feature matching degree, context matching degree and functional matching degree between the at least one entity.

[0143] Optionally, in some embodiments, the acquisition module 100 is further configured to: identify whether there is an entity among at least one entity whose multi-feature similarity is greater than a preset similarity based on the multi-feature similarity between at least one entity; if there is an entity among at least one entity whose multi-feature similarity is greater than the preset similarity, then determine that there is the same entity among at least one entity.

[0144] Optionally, in some embodiments, the acquisition module 100 is further configured to: identify the entity type and attributes of the target entity in the multi-source data, and identify the connection relationship type and attributes of the physical connection relationship between the target entities.

[0145] Optionally, in some embodiments, after generating the risk seeds of the connection relationships between failure risk entities, the risk identification module 200 is further configured to: semantically enhance the risk seeds of the connection relationships between failure risk entities using a preset vector library and a preset knowledge graph, and generate a natural language risk description.

[0146] Optionally, in some embodiments, before semantically enhancing the risk seeds of the connection relationships between failure risk entities using a preset vector library and a preset knowledge graph to generate a natural language risk description, the risk identification module 200 is further configured to: construct a knowledge graph including entity-physical connection relationship-attribute based on the entity type and attributes of the target entity, the connection relationship type of the physical connection relationship between the target entities, and the attributes of the physical connection relationship; generate embedding vectors based on the text corpus in the multi-source data, and establish an association index between the embedding vectors and the entities in the knowledge graph to obtain a vector library.

[0147] Optionally, in some embodiments, the risk assessment module 300 is further configured to: assess the failure impact score of the failure risk entity connection relationship based on a first evaluation index, assess the reliability score of the failure risk entity connection relationship based on a second evaluation index, and assess the failure impact range score of the failure risk entity connection relationship based on a third evaluation index; calculate the failure risk value of the failure risk entity connection relationship based on the failure impact score, reliability score, and failure impact range score; and determine the failure risk level of the failure risk entity connection relationship based on the failure risk value of the failure risk entity connection relationship.

[0148] Optionally, in some embodiments, after obtaining the failure risk level of the connection relationship of failure risk entities based on the failure risk assessment based on the risk seed, the risk assessment module 300 is further configured to: generate a list of failure risk entity pairs according to the failure risk level of the connection relationship of failure risk entities, generate a failure risk propagation path chain according to the connection relationship of failure risk entities, and generate a connection risk analysis report according to the connection relationship of failure risk entities; and push the list of failure risk entity pairs, the failure risk propagation path chain, and the connection risk analysis report to the target terminal.

[0149] It should be noted that the foregoing explanation of the risk identification method embodiment based on entity connection relationship also applies to the risk identification system based on entity connection relationship in this embodiment, and will not be repeated here.

[0150] This application also provides an electronic device, please refer to... Figure 5 The electronic device 20 includes a processor 501 and a memory 502. The memory 501 is used to store computer programs, and the processor 502 is used to execute the programs stored in the memory 501 to implement the risk identification method based on entity connection relationship described in any embodiment of this application.

[0151] This application also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the risk identification method based on entity connection relationships described in any embodiment of this application.

[0152] In this application, "multiple" refers to two or more.

[0153] In this application, unless otherwise expressly defined, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0154] The terms “first,” “second,” “third,” “fourth,” etc., in this application (if present) are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0155] In this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, in this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0156] Unless otherwise specified, all steps in this application may be performed sequentially or randomly. For example, if a method includes steps A and B, it means that the method may include steps A and B performed sequentially, or it may include steps B and A performed sequentially. For example, if a method may also include step C, it means that step C may be added to the method in any order. For example, the method may include steps A, B, and C, or it may include steps A, C, and B, or it may include steps C, A, and B, etc.

[0157] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A risk identification method based on entity connection relationship, characterized in that, Includes the following steps: Acquire multi-source data of the physical interface of the mechanical system, and identify target entity information and target entity connection relationship information in the multi-source data; Based on the target entity information and the target entity connection relationship information, semantic similarity matching is performed with a predefined failure rule base to obtain a semantic similarity score. When the semantic similarity score is greater than a preset threshold, a risk seed for the connection relationship of the failure risk entity is generated. Based on the risk seeds, a failure risk assessment is performed to obtain the failure risk level of the connection relationship between the failure risk entities.

2. The method of claim 1, wherein, Before identifying target entity information and target entity connection relationship information in the multi-source data, the process includes: Extract at least one entity from the multi-source data; Calculate the multi-feature similarity degree between the at least one entity, and determine whether there is a common entity among the at least one entity based on the multi-feature similarity degree between the at least one entity; If the same entity exists among the at least one entity, then the same entity is merged and deduplicated to obtain the target entity in the multi-source data.

3. The method of claim 2, wherein, The calculation of the multi-feature similarity between the at least one entity includes: Compare the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree among the at least one entity; The multi-feature similarity between the at least one entity is calculated based on the identifier matching degree, geometric feature matching degree, context matching degree, and functional matching degree between the at least one entity.

4. The method of claim 2, wherein, The step of determining whether there is a common entity among the at least one entities based on the multi-feature similarity between the at least one entity includes: Based on the multi-feature similarity between the at least one entity, identify whether there is an entity among the at least one entities whose multi-feature similarity is greater than a preset similarity; If there is an entity among the at least one entity whose multi-feature similarity is greater than a preset similarity, then it is determined that the same entity exists among the at least one entity.

5. The method of claim 1, wherein, The identification of target entity information and target entity connection relationship information in the multi-source data includes: Identify the entity type and attributes of the target entities in the multi-source data, and identify the connection relationship type and attributes of the physical connection relationships between the target entities.

6. The method of claim 1, wherein, After generating the risk seed for the connection relationships between entities with failure risks, the following is included: Using a pre-defined vector library and a pre-defined knowledge graph, the risk seeds of the failure risk entity connection relationships are semantically enhanced to generate a natural language risk description.

7. The method of claim 6, wherein, Before generating a natural language risk description by semantically enhancing the risk seeds of the failure risk entity connection relationships using a preset vector library and a preset knowledge graph, the process includes: A knowledge graph comprising entity-physical connection relationship-attribute is constructed based on the entity type and attributes of the target entities, the connection relationship type of the physical connection relationship between the target entities, and the attributes of the physical connection relationship. An embedding vector is generated based on the text corpus in the multi-source data, and an association index is established between the embedding vector and the entities in the knowledge graph to obtain the vector library.

8. The method according to claim 1, characterized in that, The failure risk level obtained by assessing the failure risk based on the risk seed and determining the connection relationship of the failure risk entities includes: The failure impact score of the connection relationship of the entities at risk of failure is evaluated based on the first evaluation index, the reliability score of the connection relationship of the entities at risk of failure is evaluated based on the second evaluation index, and the failure impact range score of the connection relationship of the entities at risk of failure is evaluated based on the third evaluation index. The failure risk value of the failure risk entity connection relationship is calculated based on the failure impact degree score, the reliability score, and the failure impact range score. The failure risk level of the failure risk entity connection relationship is determined based on the failure risk value of the failure risk entity connection relationship.

9. The method according to claim 1, characterized in that, After obtaining the failure risk level of the failure risk entity connection relationship based on the failure risk seed through failure risk assessment, the process includes: A list of failure risk entity pairs is generated based on the failure risk level of the failure risk entity connection relationship, a failure risk propagation path chain is generated based on the failure risk entity connection relationship, and a connection risk analysis report is generated based on the failure risk entity connection relationship. The failure risk entity pair list, the failure risk propagation path chain, and the connection risk analysis report are pushed to the target terminal.

10. An electronic device, characterized in that, Including processor and memory, among which Memory, used to store computer programs; A processor is used to execute a program stored in memory to implement the risk identification method based on entity connection relationship as described in any one of claims 1-9.

Images