Urban rail signal system fault diagnosis method based on knowledge graph and large model

Abstract

The embodiment of the invention provides an urban rail signal system fault diagnosis method based on a knowledge graph and a large model, and the method comprises the steps: collecting system operation logs, state parameters and fault information in real time, and carrying out the data preprocessing and standardization; based on historical fault data, a fault classification and prediction model is constructed through feature extraction and mode recognition, and automatic fault recognition and risk early warning are achieved; constructing a fault diagnosis knowledge graph, and establishing an association relationship among entities such as equipment, faults, reasons, maintenance schemes and the like; a pre-training large language model and a LoRA technology are adopted for efficient fine tuning, a fault diagnosis reasoning model is trained, and end-to-end generation from fault description to diagnosis and maintenance suggestions is achieved; the knowledge graph and large model output are fused, real-time and historical data are combined, multi-path fault analysis and comprehensive diagnosis are carried out, and a diagnosis report is generated and displayed. The intelligent level, accuracy and efficiency of fault diagnosis can be improved, and technical support is provided for intelligent operation and maintenance management.

Description

TECHNICAL FIELD

[0001] The present document relates to the field of fault diagnosis, and in particular to a method for diagnosing faults in urban rail signal systems based on knowledge graphs and large models. BACKGROUND

[0002] The CBTC system of urban rail transit, as the core control system of modern rail transit, its safety and reliability are directly related to the operation safety of the entire rail transit system. The CBTC system is mainly composed of regional controllers and on-board controllers and other core devices. These devices generate a large amount of log data during operation.

[0003] Currently, the fault diagnosis of urban rail signal systems mainly has the following problems: low efficiency of log data collection: traditional log collection methods mainly rely on manual on-site collection, which is inefficient and cannot achieve real-time monitoring; low degree of intelligent fault diagnosis: existing fault diagnosis methods mainly rely on expert experience and simple rule matching, lack intelligent analysis means, and the diagnosis effect is limited; imperfect fault knowledge management: fault diagnosis knowledge is scattered among experts and documents, lacking systematic knowledge management and sharing mechanism; lack of predictive maintenance capability: existing systems mainly diagnose after the fact, lacking the ability to predict potential faults.

[0004] There are already some fault diagnosis technical solutions based on knowledge graphs, but there are obvious limitations: CN116484014A patent proposes a fault display method based on knowledge graph, which constructs a knowledge graph according to historical fault data and displays fault information in a tree structure. Although this solution can reduce the difficulty of fault troubleshooting, it has the following shortcomings: limited to fault information retrieval and display, lacking intelligent fault diagnosis reasoning ability; does not involve real-time data collection and processing, cannot support dynamic fault monitoring; lacks predictive maintenance function, still belongs to passive fault handling; CN118656427A patent provides a signal intelligent operation and maintenance fault retrieval system, which uses a distributed search and analysis engine to query fault data. This solution has improved data processing, but still has limitations: mainly focuses on fault data storage and retrieval, lacks deep fault analysis capability; does not use advanced AI technologies such as large models, the degree of intelligence is limited; lacks semantic reasoning ability of knowledge graph, cannot perform complex correlation analysis.

[0005] Therefore, there is an urgent need for a comprehensive technical solution that can integrate knowledge graphs, large models and other advanced technologies to achieve automated log collection, intelligent fault diagnosis and knowledge-based management. SUMMARY

[0006] One or more embodiments of the specification provide a knowledge graph and large model-based urban rail signal system fault diagnosis method, comprising: S1: Real-time collection of urban rail signal system operation logs, state parameters and fault alarm information, and preprocessing of the collected data to form a unified and standardized data set; S2: Statistical analysis, feature extraction and pattern recognition of historical fault data, construction of fault classification model and prediction model, and realization of automatic identification of fault mode and risk warning; S3: Construction of fault diagnosis knowledge graph, establishment of the correlation network between equipment, fault phenomenon, fault cause, maintenance scheme, technical parameter and maintenance history through entity recognition, relationship extraction and graph storage, and formation of a complete fault diagnosis knowledge system; S4: Based on the pre-trained large language model, combining LoRA low-rank adaptation technology for efficient fine-tuning and distributed training of parameters, constructing an intelligent inference model suitable for fault diagnosis, and realizing end-to-end generation from fault description to diagnosis result and maintenance suggestion; S5: Fusion of knowledge graph reasoning and large language model output, combination of real-time monitoring data and historical experience, multi-path fault analysis and comprehensive diagnosis, generation of a diagnosis report containing fault type, cause analysis and maintenance scheme, and real-time output and display.

[0007] Further, the equipment for data collection includes regional controllers and on-board controllers in the urban rail signal system; The specific method for preprocessing the collected data to form a unified and standardized data set includes: Data cleaning to remove duplicate logs, invalid data and noise data; Processing of abnormal values in the data, using isolated forest or Z-score algorithm to identify and process abnormal data points; Using mean filling, interpolation method or machine learning-based prediction filling method to fill in missing values; Converting log data from different sources into a unified JSON or XML format; Using Min-Max standardization or Z-score standardization to normalize numerical features; Using TLS / SSL protocol to encrypt log data in transmission.

[0008] Further, the statistical analysis, feature extraction and pattern recognition of historical fault data, and the construction of fault classification model and prediction model specifically include: Deep analysis of the collected historical fault data to extract key information such as time, location, equipment type, fault phenomenon and processing process of fault occurrence; Based on the analysis results of fault data, the faults are systematically classified according to the dimensions of equipment type, fault type, severity, and influence range; For each fault classification, the typical characteristics, occurrence conditions, fault signs, and treatment methods of the faults are analyzed in depth; Based on the characteristic analysis of the classified faults, the analysis rules for fault diagnosis and treatment are automatically mined using machine learning and expert system technology; The mined fault analysis rules are verified for accuracy and effectiveness; The fault analysis rules that pass the verification are stored in the fault rule library in a standardized format, and a version management and updating mechanism for the rules is established.

[0009] Further, the fault diagnosis knowledge system construction method is as follows: Collect historical fault data, expert knowledge, technical documents, and maintenance records as raw data, and perform data cleaning, format standardization, noise filtering, and data verification processing on the raw data; Extract entities from the processed raw data, including equipment entities, fault phenomena, maintenance schemes, and technical parameters, and use rule-based matching, conditional random fields, or BERT models for named entity recognition; Extract equipment-fault association, fault-maintenance association, and parameter-range association, and construct a weighted directed graph; Use Neo4j graph database to store the entity relationship network and support Cypher query language for multi-hop query and path analysis; Establish an incremental updating and version control mechanism to periodically integrate newly generated fault cases and expert experience knowledge through incremental learning.

[0010] Further, step S4 specifically includes: Obtain real-time equipment operation logs, historical fault records, and expert maintenance experience multi-source heterogeneous data, convert the collected data into instruction-input-output triple format suitable for large language model processing, and perform data quality control; Configure GPU clusters, deep learning frameworks, and other infrastructure, select a pre-trained large language model suitable for fault diagnosis tasks as a pre-trained base model, use low-rank adaptation LoRA or quantized low-rank adaptation QLoRA technology for efficient fine-tuning of parameters, configure training hyperparameters, use a distributed training framework for efficient parallel training, and monitor the training process in real time to ensure model convergence; Combine the fine-tuned adapter with the base model to build a specialized inference model, and realize end-to-end generation of fault description to diagnosis results and maintenance suggestions.

[0011] Further, step S5 specifically includes: Rule reasoning based on knowledge graph, matching existing fault modes and treatment schemes; Outputting fault probability and confidence score using machine learning model; Introducing large language model for fault semantic analysis and generative inference; Confidence weighted fusion of the above multi-source reasoning results, outputting comprehensive diagnostic report including fault type, possible cause, repair step, estimated processing time and required spare parts information.

[0012] Further, the system operation log collection supports collection of multiple log formats and communication protocols, including ATP logs, ATO logs and ZC logs.

[0013] Further, the method further includes a fault warning step: By real-time monitoring of equipment state parameters and running trend, combined with prediction model and threshold rule, identifying potential fault risk and generating graded warning information.

[0014] Further, the method further includes a repair effect evaluation: Continuous monitoring and effect evaluation of the repaired equipment, collecting equipment operation data, comparing key indicators before and after repair, evaluating repair effect, adding successful repair cases to training set and knowledge graph, optimizing diagnostic model and graph structure.

[0015] One or more embodiments of the present specification provide a city rail signal system fault diagnosis system based on knowledge graph and large model, comprising: Data collection module: for real-time collection of city rail signal system operation log, state parameter and fault alarm information, and pre-processing of the collected data to form a unified and standardized data set; Historical fault data analysis module: for statistical analysis, feature extraction and pattern recognition of historical fault data, building fault classification model and prediction model, realizing automatic identification and risk warning of fault mode; Knowledge graph construction module: for constructing fault diagnosis knowledge graph, establishing the correlation network between equipment, fault phenomenon, fault cause, repair scheme, technical parameter and repair history through entity recognition, relationship extraction and graph storage, forming a complete fault diagnosis knowledge system; Fault diagnosis module: for parameter efficient fine-tuning and distributed training based on pre-trained large language model combined with LoRA low-rank adaptation technology, constructing intelligent inference model suitable for fault diagnosis, realizing end-to-end generation from fault description to diagnosis result and repair suggestion; The result output module is used for fusing knowledge graph reasoning and large language model output, combining real-time monitoring data and historical experience to perform multi-path fault analysis and comprehensive diagnosis, generate a diagnosis report containing fault types, cause analysis and maintenance schemes, and perform real-time output display.

[0016] By adopting the embodiment of the present application, the log collection efficiency is significantly improved: through the remote automatic collection technology of the four-layer distributed architecture, real-time collection and high-quality preprocessing of device operation logs are realized, the reliability of data transmission and the efficiency of log preprocessing are significantly improved, and the low-efficiency mode of traditional manual field collection is completely changed; the intelligent level of fault diagnosis is greatly improved: the intelligent reasoning algorithm based on large model fine-tuning combined with knowledge graph technology significantly improves the fault diagnosis accuracy and response speed, realizing the fundamental change from expert experience dependence to intelligent automatic diagnosis; the systematic management of fault knowledge is realized: a complete knowledge graph containing device entities, fault phenomena, maintenance schemes, technical parameters and other multi-dimensional information is constructed, realizing the structured storage, intelligent retrieval and dynamic update of fault diagnosis knowledge; the predictive maintenance capability is significantly enhanced: through the deep integration of machine learning models and knowledge graphs, potential fault patterns can be effectively identified, the predictive maintenance capability is significantly improved, and the device failure rate and maintenance cost are effectively reduced; a full-stack integrated solution is provided: an integrated intelligent diagnosis platform based on a five-layer micro-service architecture is developed, supporting large-scale data processing, realizing the full-process automation from data collection, knowledge construction, intelligent reasoning to visualization display, and providing a complete intelligent fault diagnosis solution for urban rail signal systems; a continuous learning optimization mechanism is established: through incremental learning and version control mechanism, the system can automatically learn new fault cases, continuously optimize the diagnosis model and knowledge graph, and ensure the continuous improvement of diagnosis capability and the continuous evolution of technical solutions.

[0017] The above description is only a summary of the technical scheme of the present application. In order to more clearly understand the technical means of the present application, the following specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical scheme in the one or more embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only some embodiments described in the present specification, and those skilled in the art can also obtain other drawings according to these drawings without paying creative labor.

[0019] Figure 1A flowchart of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification; Figure 2 A data collection flowchart of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification; Figure 3 A historical fault data analysis flowchart of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification; Figure 4 A fault diagnosis knowledge system construction flowchart of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification; Figure 5 An intelligent fault diagnosis reasoning algorithm flowchart of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification; Figure 6 An integrated intelligent diagnosis platform architecture diagram of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification; Figure 7 A composition schematic diagram of a knowledge graph and large model based urban rail signal system fault diagnosis system provided for one or more embodiments of the present specification. DETAILED DESCRIPTION

[0020] In order to enable persons skilled in the art to better understand the technical solutions in one or more embodiments of the present specification, the technical solutions in one or more embodiments of the present specification will be described clearly and completely below in conjunction with the accompanying drawings of the present specification. Obviously, the described embodiments are only a part of the embodiments of the present specification, rather than all the embodiments. Based on one or more embodiments of the present specification, all other embodiments obtained by persons skilled in the art without creative labor should belong to the protection scope of the present document.

[0021] Method embodiments According to the embodiments of the present application, a knowledge graph and large model based urban rail signal system fault diagnosis method is provided, Figure 1 A flowchart of a knowledge graph and large model based urban rail signal system fault diagnosis method provided for one or more embodiments of the present specification is shown as Figure 1 The knowledge graph and large model based urban rail signal system fault diagnosis method according to the embodiments of the present application specifically includes: S1: Real-time collection of urban rail signal system operation logs, state parameters and fault alarm information, and preprocessing of the collected raw log data to form a unified and standardized data set.

[0022] The device for data collection includes a regional controller and a vehicle-mounted controller in the urban rail signal system, and the data collection process is as shown in Figure 2 S101. Regional controller data collection: Real-time collection of system operation logs, device state logs, and communication logs, monitoring of the running state and performance parameters of the regional controller, comprehensive monitoring of the running state of the ZC device through the data collection agent deployed on the regional controller, including detailed records of each stage such as device startup, operation, failure, and maintenance; S102. Vehicle-mounted controller data collection: Collection of vehicle-mounted ATP logs, train operation logs, and communication state logs, tracking of train running trajectory and control instruction execution, real-time collection and recording of key parameters such as speed, position, braking, and traction during train operation through the vehicle-mounted data collection terminal; S103. ZC collection agent: Responsible for log file monitoring, real-time data grabbing, and local cache management, ensuring the continuity and integrity of data collection, using multi-threading technology to achieve concurrent monitoring of multiple log files, and capturing newly added log content in a timely manner through file change detection mechanism; S104. CC collection agent: Realizing protocol analysis, data format conversion, and transmission queue management, supporting unified processing of multiple data formats, built-in multiple communication protocol analyzers to handle data formats of different manufacturers' devices and convert them into standardized data structures; S105. Data preprocessing: Performing data cleaning, format standardization, quality inspection, and abnormality filtering, etc. to ensure data quality, specifically: Data cleaning to remove duplicate logs, invalid data, and noise data; processing of abnormal values in the data using Isolation Forest or Z-score algorithm to identify and process abnormal data points; using mean filling, interpolation method, or machine learning-based prediction filling method to fill in missing values; converting log data from different sources into unified JSON or XML format; using Min-Max standardization or Z-score standardization to normalize numerical value features; using TLS / SSL protocol to encrypt log data during transmission; S106. Central database: Providing structured storage, index optimization, backup mechanism, and query interface, supporting efficient storage and retrieval of massive data, and the database adopts distributed architecture design, supporting horizontal expansion and being able to handle PB-level historical data storage.

[0023] ​The system operation log collection supports collection of multiple log formats and communication protocols, including ATP logs, ATO logs and ZC logs.

[0024] S2: Statistical analysis, feature extraction and pattern recognition are performed on historical failure data to construct a failure classification model and a prediction model, and automatic identification of failure modes and risk early warning are realized.

[0025] The historical failure data analysis flowchart is shown in Figure 3 S201. Historical failure data analysis: In-depth analysis is performed on the collected historical failure data to extract key information such as the time, location, equipment type, failure phenomenon, and treatment process of the failure, and data mining techniques are used to identify the rules and characteristics of failure occurrence from massive historical data; S202. Failure classification and induction: Based on the failure data analysis results, the failure is systematically classified according to dimensions such as equipment type, failure type, severity, and impact range, and a multi-level failure classification system is established to provide a classification basis for subsequent rule mining; S203. Classification failure feature extraction: For each failure classification, the typical features, occurrence conditions, failure signs and treatment methods of the failure are analyzed in depth, and pattern recognition algorithms are used to extract common features and differentiated features of various failures; S204. Failure analysis rule mining: Based on the feature analysis of the classified failure, machine learning and expert system techniques are used to automatically mine the analysis rules for failure diagnosis and treatment, decision tree, association rule mining and other algorithms are used to generate "if-then" form of failure analysis rules; S205. Rule verification and optimization: The accuracy and effectiveness of the mined failure analysis rules are verified, the reliability of the rules is ensured through cross-validation and expert review, a rule quality evaluation mechanism is established, and the accuracy and applicability of the rules are continuously optimized; S206. Failure rule storage management: The verified failure analysis rules are stored in the failure rule library according to the standardized format, a rule version management and updating mechanism is established, and the rules are classified, quickly searched and dynamically updated to provide rule support for intelligent failure diagnosis.

[0026] S3: A failure diagnosis knowledge graph is constructed, an association network between equipment, failure phenomena, failure causes, repair schemes, technical parameters and repair history is established through entity recognition, relationship extraction and graph storage, and a complete failure diagnosis knowledge system is formed.

[0027] The failure diagnosis knowledge system construction method is shown in Figure 4 Stage 1: Data preparation and preprocessing ​​S301. Data source: Collect historical failure data, expert knowledge, technical documents, maintenance records, and other multi-source data as raw data, establish a comprehensive knowledge source system, and ensure the integrity and authority of the knowledge source.

[0028] S302. Data preprocessing: Data cleaning, format standardization, noise filtering, data verification, and other processing are performed on the raw data to ensure data quality and consistency. Data cleaning algorithms are used to automatically identify and handle data quality issues. Stage 2: Entity recognition and extraction Extract entities from processed raw data, including device entities, failure phenomena, maintenance solutions, and technical parameters. Use rule-based matching, conditional random fields, or BERT models for named entity recognition. S303. Device entity: Identify regional controller ZC, vehicle controller CC, signal machine, switch machine, track circuit, axle counter, and other device names, model specifications, and technical parameters. Use named entity recognition technology to accurately identify device entities. S304. Failure phenomenon: Extract failure types, symptom descriptions, and severity levels such as communication interruption, signal loss, speed anomaly, and position deviation. Use text mining techniques to automatically identify key information in failure descriptions. S305. Maintenance solution: Extract processing steps, tools and materials, and operation duration such as restarting devices, replacing components, software upgrades, and parameter adjustments. Use pattern recognition algorithms to extract standardized maintenance processes from maintenance records. S306. Technical parameter: Identify technical parameters such as operating speed, communication delay, signal strength, voltage and current, normal range, and monitoring standards. Establish a parameter identification model to automatically extract key parameter information from technical documents. Stage 3: Relationship extraction and establishment Extract device-failure association, failure-maintenance association, and parameter-range association, and construct a weighted directed graph. S307. Device-failure association: Establish the association between devices and failure phenomena, and statistically analyze failure frequency, failure mode, and impact range. Use association rule mining algorithms to discover statistical associations between devices and failures. S308. Failure-maintenance association: Establish the correspondence between failure phenomena and maintenance solutions, analyze maintenance effectiveness, maintenance duration, and resource consumption, and establish a failure-maintenance mapping model to provide maintenance decision support for fault diagnosis. S309. Parameter-range association: Define the normal operating range and abnormal judgment criteria for technical parameters, establish an early warning mechanism, determine the normal range of parameters through statistical analysis, and construct abnormal detection rules. Stage 4: Knowledge graph construction and storage S310. Graph database storage: Build entity relationship network using Neo4j graph database, realize node storage, relationship storage, index optimization, query optimization and other functions, support multi-hop query and path analysis with Cypher query language, realize efficient storage management of large-scale knowledge graph by deploying distributed graph database cluster; Stage 5: Knowledge update and maintenance S311. Knowledge update mechanism: Establish incremental update and version control mechanism, support automatic learning of new fault cases and dynamic update of knowledge graph, build streaming data processing architecture, realize continuous optimization and version management of knowledge graph.

[0029] S4: Based on pre-trained large language model, combined with LoRA low rank adaptation technology for parameter efficient fine-tuning and distributed training, build intelligent reasoning model suitable for fault diagnosis, realize end-to-end generation from fault description to diagnosis result and maintenance suggestion.

[0030] The intelligent fault diagnosis reasoning algorithm flow chart is as shown in Figure 5 S401. Data preparation: Obtain real-time device running log, historical fault record, expert maintenance experience and other multi-source heterogeneous data, convert the collected structured and unstructured data into instruction-input-output triple format suitable for large language model processing, and perform data quality control to ensure high-quality training data, establish a comprehensive fault diagnosis data foundation, and provide high-quality data support for subsequent model training; S402. Model training: Configure GPU cluster, deep learning framework and other infrastructure, select a pre-trained large language model suitable for fault diagnosis task as a pre-trained base model, use low rank adaptation LoRA or quantized low rank adaptation QLoRA technology for parameter efficient fine-tuning, configure appropriate training hyperparameters, use distributed training framework to realize efficient parallel training, monitor the training process in real time to ensure model convergence, and make the model have professional ability of fault diagnosis through professional fine-tuning; S403. Fault reasoning: Combine the fine-tuned adapter with the base model to build a specialized reasoning model, design professional prompt word templates to guide the model to generate high-quality diagnosis results, use the fine-tuned model to generate fault diagnosis analysis, cause identification, maintenance suggestions and other content based on the input fault information, realize intelligent fault diagnosis reasoning, and provide professional diagnosis suggestions; ​S404. Diagnosis result: The generated diagnosis text is structured to form a standardized diagnosis report containing information such as fault type, severity, possible causes, repair scheme, and estimated duration. The trained intelligent diagnosis model is integrated into the overall fault diagnosis system, and a continuous learning mechanism is established to realize the self-evolution of the system, ensuring the standardization and operability of the diagnosis results, and building a complete intelligent diagnosis solution.

[0031] S5: Fusion of knowledge graph reasoning and large language model output, combining real-time monitoring data and historical experience, multi-path fault analysis and comprehensive diagnosis, generating diagnosis report containing fault type, cause analysis and repair scheme, and real-time output display.

[0032] Specifically, based on knowledge graph, rule reasoning is performed to match existing fault modes and processing schemes; machine learning model is used to output fault probability and confidence score; large language model is introduced for fault semantic analysis and generative inference; the above multi-source reasoning results are weighted and fused according to confidence, and a comprehensive diagnosis report is output, including fault type, possible cause, repair step, estimated processing time and required spare parts information.

[0033] The method further includes a fault warning step: by monitoring equipment state parameters and running trend in real time, combining prediction model and threshold rule, identifying potential fault risk and generating graded warning information.

[0034] The method further includes a repair effect evaluation: continuously monitoring and evaluating the effect of the repaired equipment, collecting equipment operation data, comparing key indicators before and after repair, evaluating repair effect, adding successful repair cases to training set and knowledge graph, optimizing diagnosis model and graph structure.

[0035] By developing an integrated intelligent diagnosis platform integrating data processing, model prediction, knowledge reasoning and visualization display, the whole process of fault diagnosis is realized automatically, intelligently and visually, and the platform includes five layers of design, as shown in Figure 6 Presentation layer: S501. Web management interface: provides system management, user interface and other functions, supports multi-role permission control, builds a web-based management console, and realizes visual management and operation of the system.

[0036] S502. API interface service: provides RESTful API, data interface and other standardized interfaces, supports interconnection between systems, designs standardized API interfaces, and realizes seamless integration with external systems.

[0037] ​S503. Report system: Realize statistical report, data analysis and other functions, support multi-dimensional data analysis, build rich data visualization components, provide intuitive data analysis and display capabilities.

[0038] Business logic layer: S504. Fault diagnosis engine: Integrate inference algorithms, diagnostic logic and other core diagnostic functions, encapsulate core diagnostic algorithms, and provide standardized fault diagnosis service interfaces.

[0039] S505. Knowledge management module: Manage knowledge graph, rule management and other knowledge assets, realize the whole life cycle management of knowledge graph, and ensure the continuous update and optimization of knowledge.

[0040] S506. Workflow management: Provide process control, task scheduling and other process management, establish a standardized business process management mechanism, and improve work efficiency and standardization.

[0041] Service layer: S507. Data collection service: Responsible for real-time collection, data cleaning, format conversion and other data processing, build a unified data collection and processing platform, and ensure the quality and consistency of data.

[0042] S508. Graph database service: Support knowledge storage, graph query, relationship analysis and other graph computing functions, deploy professional graph database, and optimize the storage and query performance of knowledge graph.

[0043] S509. Message queue service: Realize asynchronous processing, event-driven, decoupling communication and other message mechanisms, establish a reliable message transmission mechanism, and improve the responsiveness and scalability of the system.

[0044] S510. Cache service: Provide data caching, performance optimization, fast access and other acceleration functions, deploy distributed cache system, and improve data access performance and user experience.

[0045] Data layer: S511. Relational database: Store business data, user information, configuration data and other structured data, build a highly available relational database cluster, and ensure reliable storage of business data.

[0046] S512. Graph database: Manage knowledge graph, relationship network, semantic data and other graph structure data, establish a special graph data storage system, and support efficient management of complex relationship data.

[0047] S513. Time series database: Process monitoring data, historical records, trend analysis and other time series data, establish a time series data storage system, and support efficient processing of large-scale time series data.

[0048] S514. Document database: Store unstructured data, document storage, full-text search, and other document data, build a document data management system, support storage and retrieval of multiple formats of documents.

[0049] Security and monitoring layer: S515. Identity authentication: Implement user login, single sign-on, multi-factor authentication, and other security authentication, establish a unified identity authentication system, and ensure the security verification of user identity.

[0050] S516. Permission control: Provide role management, access control, operation authorization, and other permission management, build a role-based access control mechanism, and ensure the safe access of system resources.

[0051] S517. Audit log: Record operation records, security audit, compliance check, and other audit functions, establish a complete audit log system, and ensure the traceability of system operation.

[0052] S518. System monitoring: Implement performance monitoring, health check, alarm mechanism, and other operation and maintenance monitoring, establish a comprehensive system monitoring system, and real-time grasp the system running state.

[0053] The beneficial effects of the present application are as follows: Significantly improve the efficiency of log collection: Through the remote automatic collection technology of four-layer distributed architecture, the real-time collection and high-quality preprocessing of device operation logs are realized, the reliability of data transmission and the efficiency of log preprocessing are significantly improved, and the traditional inefficient mode of manual field collection is completely changed; greatly improve the intelligent level of fault diagnosis: based on the intelligent reasoning algorithm of large model fine-tuning combined with knowledge graph technology, the fault diagnosis accuracy and response speed are significantly improved, realizing the fundamental change from expert experience dependence to intelligent automatic diagnosis; realize the systematic management of fault knowledge: a complete knowledge graph containing device entities, fault phenomena, maintenance schemes, technical parameters and other multi-dimensional is constructed, realizing the structured storage, intelligent retrieval and dynamic update of fault diagnosis knowledge; significantly enhance the predictive maintenance capability: through the deep integration of machine learning model and knowledge graph, potential fault patterns can be effectively identified, the predictive maintenance capability is significantly improved, and the equipment failure rate and maintenance cost are effectively reduced; provide full-stack integrated solution: develop an integrated intelligent diagnosis platform based on five-layer micro-service architecture, support large-scale data processing, realize the full-process automation from data collection, knowledge construction, intelligent reasoning to visualization display, and provide a complete intelligent fault diagnosis solution for urban rail signal systems; establish a continuous learning optimization mechanism: through incremental learning and version control mechanism, the system can automatically learn new fault cases, continuously optimize the diagnosis model and knowledge graph, and ensure the continuous improvement of diagnosis capability and the continuous evolution of technical scheme.

[0054] System embodiment According to the embodiment of the present application, a knowledge graph and large model-based urban rail signal system fault diagnosis system is provided, Figure 7 A composition schematic diagram of a knowledge graph and large model-based urban rail signal system fault diagnosis system provided for one or more embodiments of the present specification is shown in Figure 7 As shown, the knowledge graph and large model-based urban rail signal system fault diagnosis system according to the embodiment of the present application specifically comprises: The data acquisition module is used for real-time acquisition of urban rail signal system operation logs, state parameters and fault alarm information, and pre-processing of the collected raw log data to form a unified and standardized data set. The historical fault data analysis module is used for statistical analysis, feature extraction and pattern recognition of historical fault data, construction of fault classification models and prediction models, and realization of automatic identification and risk warning of fault patterns. The knowledge graph construction module is used for constructing a fault diagnosis knowledge graph, establishing an association network between equipment, fault phenomena, fault causes, maintenance schemes, technical parameters and maintenance history through entity recognition, relationship extraction and graph storage, and forming a complete fault diagnosis knowledge system. The fault diagnosis module is used for efficient fine-tuning and distributed training of parameters based on a pre-trained large language model combined with LoRA low-rank adaptation technology, construction of an intelligent inference model suitable for fault diagnosis, and realization of end-to-end generation of fault description to diagnosis results and maintenance suggestions. The result output module is used for fusion of knowledge graph reasoning and large language model output, combination of real-time monitoring data and historical experience, multi-path fault analysis and comprehensive diagnosis, generation of a diagnosis report containing fault types, cause analysis and maintenance schemes, and real-time output and display.

[0055] The embodiment of the present application is a system embodiment corresponding to the above-mentioned method embodiment, and the specific operation of each module can be understood with reference to the description of the method embodiment, which will not be repeated here.

[0056] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the above embodiments, or make equivalent replacement to part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A knowledge graph and large model-based urban rail signal system fault diagnosis method, characterized in that, Comprise: S1: Real-time acquisition of urban rail signal system operation log, state parameter and fault alarm information, and preprocessing of the collected data to form a unified and standardized data set; S2: Statistical analysis, feature extraction and pattern recognition of historical fault data, construction of fault classification model and prediction model, and realization of automatic identification and risk warning of fault mode; S3: Construction of fault diagnosis knowledge graph, establishment of the correlation network between equipment, fault phenomenon, fault cause, maintenance scheme, technical parameter and maintenance history through entity recognition, relationship extraction and graph storage, and formation of a complete fault diagnosis knowledge system; S4: Based on the pre-trained large language model, combining LoRA low rank adaptation technology for efficient fine tuning and distributed training of parameters, constructing an intelligent reasoning model suitable for fault diagnosis, and realizing end-to-end generation from fault description to diagnosis result and maintenance suggestion; S5: Fusion of knowledge graph reasoning and large language model output, combination of real-time monitoring data and historical experience, multi-path fault analysis and comprehensive diagnosis, generation of diagnosis report containing fault type, cause analysis and maintenance scheme, and real-time output and display.

2. The method of claim 1, wherein, The equipment for data collection includes regional controllers and on-board controllers in the urban rail signal system; The specific method for preprocessing the collected data to form a unified and standardized data set includes: Data cleaning to remove duplicate logs, invalid data and noise data; Processing of abnormal values in the data, using isolated forest or Z-score algorithm to identify and process abnormal data points; Using mean filling, interpolation method or machine learning-based prediction filling method to fill missing values; Converting log data from different sources into unified JSON or XML format; Using Min-Max standardization or Z-score standardization to normalize numerical features; Using TLS / SSL protocol to encrypt log data during transmission.

3. The method of claim 1, wherein, The specific method for statistical analysis, feature extraction and pattern recognition of historical fault data, construction of fault classification model and prediction model includes: Deep analysis of collected historical fault data, extraction of key information such as fault occurrence time, location, equipment type, fault phenomenon and processing process; Based on the results of fault data analysis, systematically classifying faults according to equipment type, fault type, severity, impact range, etc.; In-depth analysis of typical features, occurrence conditions, fault signs and processing methods for each fault classification; Based on the feature analysis of classified faults, using machine learning and expert system technology to automatically mine fault diagnosis and processing analysis rules; Verifying the accuracy and effectiveness of the mined fault analysis rules; Storing the verified fault analysis rules in the fault rule library according to the standardized format, establishing a version management and updating mechanism for the rules.

4. The method of claim 1, wherein, The fault diagnosis knowledge system construction method is as follows: Collect historical fault data, expert knowledge, technical documents and maintenance record data as raw data, and perform data cleaning, format standardization, noise filtering and data verification processing on the raw data; Extracting entities from the processed raw data, including device entities, fault phenomena, repair schemes and technical parameters, using rule-based matching, conditional random fields or BERT models for named entity recognition; Extracting device-fault association, fault-repair association, parameter-range association, and constructing a weighted directed graph; Using Neo4j graph database to store entity relationship network, supporting Cypher query language for multi-hop query and path analysis; Establishing incremental update and version control mechanism, periodically integrating new fault cases and expert experience knowledge through incremental learning mechanism.

5. The method of claim 1, wherein, Step S4 specifically includes: Obtaining real-time device operation logs, historical fault records, and expert repair experience multi-source heterogeneous data, converting the collected data into instruction-input-output triple format suitable for large language model processing, and performing data quality control; Configuring GPU cluster, deep learning framework and other infrastructure, selecting a pre-trained large language model suitable for fault diagnosis task as a pre-trained base model, using low-rank adaptation LoRA or quantized low-rank adaptation QLoRA technology for efficient fine-tuning of parameters, configuring training hyperparameters, and using distributed training framework to achieve efficient parallel training, and real-time monitoring of the training process to ensure model convergence; Combining the fine-tuned adapter with the base model to build a specialized inference model, realizing end-to-end generation from fault description to diagnosis result and repair suggestion.

6. The method of claim 1, wherein, Step S5 specifically includes: Rule-based reasoning based on knowledge graph, matching existing fault patterns and treatment schemes; Using machine learning models to output fault probability and confidence score; Introducing large language models for fault semantic analysis and generative inference; Weighted fusion of the above multi-source reasoning results based on confidence, outputting a comprehensive diagnostic report including fault type, possible cause, repair steps, estimated processing time and required spare parts information.

7. The method of claim 1, wherein, The system supports the collection of multiple log formats and communication protocols during system log collection, including ATP logs, ATO logs and ZC logs.

8. The method of claim 1, wherein, The method further includes a fault warning step: By monitoring device state parameters and running trends in real time, combining prediction models and threshold rules, identifying potential fault risks and generating graded warning information.

9. The method of claim 1, wherein, The method further includes a repair effect evaluation: Continuously monitoring and evaluating the effect of the repaired device, collecting device operation data, comparing key indicators before and after repair, evaluating repair effect, adding successful repair cases to the training set and knowledge graph, and optimizing the diagnostic model and graph structure.

10. A knowledge graph and large model-based urban rail signal system fault diagnosis system, characterized in that, Including: Data acquisition module: used for real-time acquisition of urban rail signal system operation logs, state parameters and fault alarm information, and pre-processing of the collected data to form a unified and standardized data set; Historical fault data analysis module: used for statistical analysis, feature extraction and pattern recognition of historical fault data, construction of fault classification model and prediction model, and realization of automatic identification and risk warning of fault mode; Knowledge graph construction module: used for constructing fault diagnosis knowledge graph, through entity recognition, relation extraction and graph storage, establishing the correlation network between equipment, fault phenomenon, fault cause, maintenance scheme, technical parameter and maintenance history, forming a complete fault diagnosis knowledge system; Fault diagnosis module: used for parameter efficient fine-tuning and distributed training based on pre-trained large language model combined with LoRA low-rank adaptation technology, constructing intelligent reasoning model suitable for fault diagnosis, realizing end-to-end generation from fault description to diagnosis result and maintenance suggestion; Result output module: used for fusing knowledge graph reasoning and large language model output, combining real-time monitoring data and historical experience, performing multi-path fault analysis and comprehensive diagnosis, generating diagnosis report containing fault type, cause analysis and maintenance scheme, and performing real-time output display.

Images