Rail transit fault decision-making system based on enhanced local AI large model

Abstract

The invention provides a rail transit fault decision-making system based on an enhanced local AI large model, intelligent decision-making of rail transit faults is realized through cooperation of four types of modules, and a visual billboard module builds a B / S architecture interface based on Vue3 and provides fault data display and interaction functions; the data processing and storage module integrates the dispersed data, processes the data and constructs a fault data database and a rail transit knowledge base; the large model construction module is used for locally deploying an AI model and creating a business workflow and an AI application; the function interface module supports communication cooperation of all the modules. According to the method, industrial pain points such as data islands and manual decision dependence are solved, a local AI large model is supported and enhanced through prompt fine adjustment, a professional rail transit knowledge base and a fault data database, intelligence of treatment decision making of subway equipment faults is achieved, and the efficiency, the professional degree and the accuracy of fault analysis are improved through simple and easy-to-use system tools.

Description

Technical Field

[0001] This invention relates to the field of rail transit technology, and in particular to a rail transit fault decision-making system based on an enhanced local AI large model. Background Technology

[0002] In the rail transit industry, traditional operation and failure decision-making mainly relies on human experience and static analysis of historical data. However, with the expansion of system scale, the increase in equipment complexity, and the increasingly dynamic and changeable operating environment, these traditional methods have gradually revealed the following shortcomings: First, there are significant obstacles to data integration and sharing. On the one hand, the industry suffers from prominent data silos. Multi-source, heterogeneous data generated from different stages such as operation, maintenance, and scheduling are difficult to integrate and apply collaboratively due to the lack of a unified integration mechanism. This results in incomplete information available for decision-making, failing to provide comprehensive support for decision-making. On the other hand, data management is fragmented, with each department independently managing its own business data. A cross-departmental data sharing system has not been established, severely restricting the flow and reuse of data. Decision-makers struggle to obtain comprehensive and accurate data resources, further hindering the rationality of their decisions.

[0003] Secondly, the decision-making process suffers from insufficient real-time performance and dynamic adaptability. From a technical perspective, traditional data integration technologies suffer from low processing efficiency and high operating costs when dealing with the large-scale, high-frequency data generated in the rail transit industry. Furthermore, they cannot achieve real-time data updates, making it difficult for the system to perceive dynamic changes in operational data in real time. From a data processing perspective, traditional methods primarily rely on offline processing. The time lag between data collection, analysis, and final application to decision-making is considerable, failing to meet the real-time requirements of scenarios such as equipment fault monitoring and sudden passenger flow scheduling. These two factors combined result in traditional decision-making methods being unable to quickly respond to complex and ever-changing operating environments, leading to significant lags in decision-making strategy adjustments and an inability to adapt to dynamic changes in operating conditions.

[0004] Furthermore, the scientific rigor and accuracy of human decision-making are limited. Traditional decision-making relies heavily on the personal experience of staff and can only make judgments based on limited data analysis results. When faced with complex operational problems involving multiple intertwined factors, staff find it difficult to make accurate judgments in a short period of time. At the same time, the decision-making process lacks effective technical tools to support it, making it impossible to accurately predict equipment failure risks, operational optimization directions, etc., resulting in insufficient scientific rigor and accuracy in decision-making and a high risk of decision-making bias.

[0005] Finally, the knowledge management and experience transfer system is inadequate. Professional knowledge and operational experience in the rail transit industry are largely scattered across technical documents, operation manuals, and the personal experiences of veteran employees, lacking a systematic organization and management mechanism. This results in fragmented knowledge, making it difficult for new employees to quickly acquire and apply key knowledge. Furthermore, the company lacks efficient knowledge sharing and transfer channels; experience exchange among employees relies primarily on offline communication, which is not only inefficient but also prone to knowledge gaps due to staff turnover. This leads to the recurrence of previously resolved operational and equipment malfunction issues, increasing operating costs and safety risks.

[0006] Therefore, the rail transit industry urgently needs to build a new decision support system to overcome the aforementioned limitations and achieve data-driven, intelligent, real-time, and precise decision-making. In recent years, the development of artificial intelligence technology, especially the maturity of local large-scale models and knowledge base technologies, has provided new technological possibilities in this direction. Summary of the Invention

[0007] In view of the above problems, the present invention is proposed to provide a rail transit fault decision system based on an enhanced local AI large model that overcomes or at least partially solves the above problems.

[0008] This invention provides a rail transit fault decision-making system based on an enhanced local AI large model, comprising: The visualization dashboard module provides a user interface for users to interact with backend services, and visualizes and displays fault data and decision analysis information after backend service processing. The data processing and storage module is used to process rail transit fault data and rail transit knowledge, and to build a fault data database and a rail transit knowledge base based on the processed rail transit fault data and rail transit knowledge. The large model building module is used to deploy large AI models locally and create various business workflows and dialogue workflows based on the local large AI models to build various business AI applications, as well as to make fault decisions for various businesses based on rail transit fault data and rail transit knowledge using various business AI applications. The functional interface module provides various business function interfaces to support collaborative work between the data processing and storage module, the large model building module, and the visualization dashboard module, in order to realize various business functions.

[0009] Optionally, the visual dashboard module includes an AI-based real-time analysis-assisted decision-making assistant and various types of intelligent chat and dialogue toolsets.

[0010] Optionally, the data processing and storage module is also used to filter the collected rail transit operation and maintenance data to obtain rail transit fault data, preprocess the rail transit fault data, calculate rail transit fault judgment indicators based on the preprocessed rail transit fault data using various rail transit fault judgment index algorithms, and store the rail transit fault data, the preprocessed rail transit fault data, and the rail transit fault judgment indicators into the constructed database to form a fault data database.

[0011] Optionally, the data processing and storage module is also used to collect rail transit knowledge, perform basic preprocessing and sensitive data filtering on the rail transit knowledge, use segmentation methods to segment and vectorize the processed rail transit knowledge to form a rail transit knowledge base, set up retrieval methods for the rail transit knowledge base, and conduct recall tests on the rail transit knowledge base by analyzing key business requirements, and optimize the segmentation and retrieval methods of the rail transit knowledge base based on the test results to update and iterate the rail transit knowledge base.

[0012] Optionally, the business AI application built by the large model building module is used to vectorize the input rail transit fault data and rail transit knowledge using an embedded model, and to perform similarity retrieval in the rail transit knowledge base using the retrieval method of the rail transit knowledge base based on the vectorized input information. The retrieval results and the vectorized input information are then concatenated in context, and fault decisions are made based on the concatenation results.

[0013] Optionally, the visualization components of the visualization dashboard module include at least a four-quadrant chart of fault response time, a curve showing the change in lateness rate, the proportion of faulty systems and lines, a heat map of equipment inventory, equipment emergency alarms, various indicator displays, and a table of original fault data.

[0014] Optionally, the visual dashboard module is a visual webpage built using the Vue3 framework with a B / S architecture.

[0015] Optionally, a large model building module is available for building various business AI applications through the locally deployed Dify platform.

[0016] Optionally, the business AI applications built by the large model building module include at least emergency response plan applications, engineering applications, industry standard applications, quality management system applications, system requirement specification applications, business data applications, and code assistants.

[0017] Optionally, the retrieval methods of the rail transit knowledge base include at least hybrid retrieval, which includes vector retrieval and full-text retrieval.

[0018] This invention has the following advantages: This invention relates to a rail transit fault decision-making system based on an enhanced local AI large-scale model. It achieves intelligent fault decision-making through the collaborative efforts of four modules: a visualization dashboard module built on a Vue3 B / S architecture, providing fault data display and interactive functions; a data processing and storage module integrating and processing scattered data to construct a fault data database and a rail transit knowledge base; a large-scale model construction module locally deploying AI models to create business workflows and AI applications; and a functional interface module supporting communication and collaboration among the modules. This invention addresses industry pain points such as data silos and reliance on manual decision-making. Through prompts and fine-tuning, a professional rail transit knowledge base and fault data database support and enhance the local AI large-scale model, enabling intelligent decision-making for handling subway equipment faults. It uses a simple and easy-to-use system tool to improve the efficiency, professionalism, and accuracy of fault analysis. Attached Figure Description

[0019] Figure 1 This is a system overall architecture diagram provided in the embodiments of the present invention; Figure 2 This is a system technology roadmap provided in the embodiments of the present invention; Figure 3 This is a system logic flowchart provided in an embodiment of the present invention; Figure 4 This is a decision tree diagram of equipment inventory provided in an embodiment of the present invention; Figure 5 This is a flowchart of the database and knowledge base construction process provided in the embodiments of the present invention; Figure 6 This is a flowchart of the RAG questioning process provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] like Figure 1 The rail transit fault decision-making system based on an enhanced local AI large model of this invention adopts a modular design and mainly consists of the following modules: a visualization dashboard module, a data processing and storage module, a large model construction module, and various functional interface modules. The modules are efficiently connected and communicate with each other through the Spring Boot framework to ensure the real-time performance and accuracy of the data. Specifically: The visualization dashboard module, built on a B / S architecture based on the Vue3 framework, serves as the main user interface. Through interaction between the front-end and back-end services, it visually displays information such as real-time fault alarms, train delay rates, and fault system distribution in the form of charts. It also provides an AI assistant for decision support and a set of intelligent tools, integrating professional data analysis methods to provide users with a real-time monitoring and analysis interface and a faster and more accurate decision-making experience.

[0022] The data processing and storage module, after collecting real-time subway equipment fault data through business data interfaces, equipment sensors, and the Internet of Things, is responsible for integrating and processing the scattered and disorganized data within the enterprise, including subway equipment fault data. It employs data cleaning algorithms to perform data cleaning, null value imputation, outlier removal, and calculation of derived indicators, and persists the processed data to the database, thus providing reliable data support for the system, ensuring data accuracy and availability, and ultimately serving visualization and database construction. Using RAG technology, rail transit-related knowledge documents are vectorized to build a professional knowledge base, including a fault handling process knowledge base, a fault analysis experience knowledge base, and an enterprise research knowledge base. When the system needs to make fault decisions, relevant information is retrieved from the knowledge base and combined with the generative model to provide more accurate and comprehensive solutions.

[0023] The large-scale model building module, as the core business module of the system, deploys large AI models locally (such as DeepSeek and Qwen models based on VLLM and Ollama) and builds applications based on the Dify platform. This module uses prompt words for fine-tuning and combines a professional rail transit knowledge base and fault data database to enhance the large-scale model, thereby providing accurate and professional intelligent support for subway equipment fault scenarios, realizing fault analysis, value mining, and decision-making upgrades.

[0024] The functional interface module includes intelligent agent workflow interfaces, code-assisted analysis interfaces, user management interfaces, data acquisition interfaces, and an agent toolset database, responsible for forwarding various business requests and communication between modules. By connecting to the large model building module and the data processing and storage module, it provides the front-end with rich and comprehensive functional interfaces, supporting core functions such as system fault analysis, process handling, and intelligent assistance.

[0025] It can also increase automated data acquisition channels, such as business data interfaces, equipment sensors, and the Internet of Things, so that fault data can be updated in real time, further improving the system's usability and timeliness.

[0026] Reference Figure 2 System technical route and Figure 3The system logic flow and the specific implementation process of the above modules are as follows: 1. Fault Data Database Construction 1) Data Preprocessing: First, data filtering is performed. The volume of maintenance data for subway equipment across various lines is substantial. Tag filtering is used to filter non-fault-related data, such as alarm data and invalid log data. Second, standardization of formatting and enumeration is implemented. Regular expressions are used to standardize the format of data such as line names and time units. Finally, outlier handling is addressed, such as blank values. Data processing agents are used to fill in these blank values, repairing recoverable elements. Unrecoverable elements are filtered out, removing low-value, erroneous, and repetitive data to ensure safe data usage.

[0027] 2) Generation of Derivative Indicators: Decision-making requires not only atomic indicators but also algorithmic support from derived indicators, such as fault frequency density and fault impact rate, to more scientifically and intuitively demonstrate the severity of faults. Regarding equipment inventory allocation, a separate warning value is obtained by calculating the existing equipment inventory value on each line, used to determine which lines have equipment allocation imbalances requiring reallocation. The corresponding formula is as follows: Suppose there are n lines, and the equipment inventory value on each line i is S. Then the warning value T is:

[0028] Then, the front end can use a decision tree approach to determine whether a device is triggering an alarm and whether it needs to be addressed through procurement or relocation. Figure 4 As shown, if the equipment value is lower than the warning value or the equipment value is lower than the fixed value, an alarm will be triggered. If there is spare equipment of the same type, it will be allocated; otherwise, it will be procured.

[0029] 3) Database Storage: In the actual operation and maintenance management of subway equipment, a local MySQL relational database was built for efficient storage and management of various business data. This data includes equipment inventory, line faults, delay information, response time, and cost information. To ensure data integrity and consistency, the database design strictly follows the second normal form to avoid data redundancy and update anomalies.

[0030] 2. Establish a professional knowledge base for rail transit operations. like Figure 5 The steps for building the knowledge base are as follows: 1) Document collection and preprocessing First, knowledge about rail transit is collected through internal channels, including quality management systems, company emergency plans, system requirement specifications, general technical requirements, and report information. The collected documents may be in various formats, such as doc and pdf, and are uniformly converted to .md format using a tool to facilitate subsequent computer operations. CSV and .xlsx formats can retain their formatting. Then, regular expressions are used for keyword matching to filter out potentially sensitive information, such as links, email addresses, names, meaningful phone numbers, and code that can be directly executed by the front end, reducing the risk of sensitive information leakage and exposing security vulnerabilities.

[0031] 2) Text segmentation and vectorized storage The preprocessed text is segmented and divided into blocks according to segment identifiers, maximum segment lengths, and segment overlap lengths. \n\n is used as the segment identifier, and 1024 characters is used as the segment length. Consecutive spaces, newlines, and tabs are replaced. After segmentation and block division, the text is stored in the Dify knowledge base. This completes the construction of the RAG rail transit expert knowledge base.

[0032] 3) Set search method The questioning process for the RAG knowledge base is as follows: Figure 6 This approach uses embedded models to process documents, achieving more precise retrieval. A hybrid retrieval system is typically used, with vector retrieval accounting for 70% and full-text retrieval for 30%, reflecting the weighting of semantics and keywords. For example, using Tongyi Qianwen's text-embedding as a vectorization model, the input text is vectorized and then compared with the knowledge base content for similarity detection, such as using a cosine similarity algorithm to calculate the similarity between the target vector q and the knowledge base vector d.

[0033] Then, select the top K vectors with the highest similarity to the target vector from the dataset:

[0034] Simultaneously, a score threshold is set, for example, 0.4, to retain only result vectors with a similarity greater than or equal to 0.4. The search results are then obtained. Finally, this context is fed into a large AI model, such as DeepSeek, to concatenate the knowledge base results and the context, ultimately generating text output. This mechanism effectively utilizes expert knowledge bases, enhancing the output performance of AI in key tasks such as rail transit fault response, handling methods, and operation and maintenance decisions.

[0035] 4) Recall testing and management / maintenance The Dify platform allows for recall testing. Users can input keywords to view the recall results and analyze key business requirements to test the knowledge base. Further recall testing can be conducted on specific keywords to obtain approximate recall and precision rates. Based on the results, the knowledge base segmentation and retrieval methods can be optimized, with parameter adjustments or reconstruction as necessary. The knowledge base needs timely updates and iterations to reflect changes and technological advancements in the rail transit field. Since subway equipment malfunctions are also updated in real time, regular optimization and management of the knowledge base are essential to ensure its efficient operation. When necessary, a new version of the knowledge base can be rebuilt and used in subsequent Agent construction processes.

[0036] 3. Toolset for building AI intelligent agents 1) Agent Workflow Deploy the Dify large-scale model platform locally and create workflows or dialogue workflows based on business needs. Dify provides a large number of functional nodes for use. Select the LLM node, configure the local AI large-scale model (vllm-DeepSeek-R1-70B), set appropriate context or knowledge base retrieval, and fine-tune the large-scale model's output. After configuration, the node can accept input questions and automatically submit them to the large-scale model to obtain answers. Select the code execution node to execute some simple code, such as Python 3 or JavaScript code. After connecting the various nodes in sequence, external calls to the Agent can execute the process and return the answers or operation results required by the business.

[0037] 2) AI Toolset

[0038] 3) API Description Calling an Agent application requires sending the application's corresponding api_key, which is the access key. Dify will use this key information to locate the appropriate Agent. The following are generally applicable request methods: Request method: POST Request path: / chat-messages The request header 'Authorization:Bearer{api_key}' is used to transmit the api_key for authentication. Request body:

[0039] 4. Build a visual intelligent decision-making system 1) Front-end setup steps Use the Vue3 framework to build front-end pages, leveraging Vue3's component-based development features to improve development efficiency and code reusability.

[0040] Initialize the project and install necessary dependencies, such as Vue3, Echarts, dev-ui, element-plus, VueRouter, etc.

[0041] Develop components for each functional module, using the Echarts charting library to visualize the data and provide intuitive chart formats. Arrange the functional module components according to business logic and priority, employing a responsive layout design to ensure smooth information display across different devices, including but not limited to the following components:

[0042] Develop intelligent assistants and AI application toolsets, allowing users to interact with local AI models. Users can input questions into the intelligent assistant or agent using natural language, such as "Why has the failure rate increased recently?" or "How can we optimize failure response time?". The intelligent assistant combines failure data from its database and professional knowledge from its knowledge base, using natural language processing to analyze the questions and generate detailed failure analyses and decision-making suggestions. The AI ​​application toolset will, as needed, query knowledge bases such as quality management documents, project requirement documents, and emergency response plans, returning reasonable and accurate business processing suggestions.

[0043] We encapsulate an HTTP toolkit, use fetchapi to implement front-end and back-end communication, call back-end interfaces to retrieve data, and leverage the Vue3 response framework to update the content displayed on the front-end page.

[0044] Lazy loading is used for page transitions to prevent loading issues caused by frequent user switching; code splitting is used to process AI-returned messages and display the results in a more aesthetically pleasing way; timeout and error handling provide timely reminders, and timely reminders and proactive handling are provided for request failures and AI lag / timeouts caused by server issues.

[0045] 2) Backend service setup Use the Spring Boot framework to build backend services and leverage its built-in dependency management and auto-configuration features to quickly start your project.

[0046] Initialize the project and install necessary dependency packages, such as SpringWeb, SpringFlux, and SpringSecurity.

[0047] To achieve long-term data storage and efficient retrieval, Spring Data JPA is integrated with a MySQL database to enable CRUD operations and leverage JPA's caching mechanism and transaction management to optimize performance and ensure data consistency.

[0048] Based on the Agent information from the frontend request, the corresponding api_key is retrieved from the database, a new WebSocket request is generated and forwarded to the Dify interface, and then the response is returned to the frontend for display via Event information through Flux streaming. Below is an example of a Response message:

[0049] Users can also select information such as line information or file information on the page. The backend interface will read the line and files parameters and add them when generating the WebSocket request.

[0050] Besides the Agent interface development, there are still many other interfaces that need to be designed and written. The overall design adopts a RESTful API and the following interfaces have been implemented:

[0051] When presenting various types of data, we will consider using various data analysis methods and algorithms: In terms of data analysis, the Pareto principle (80 / 20 rule) is used to analyze the distribution data to identify key factors and optimize resource allocation; decision tree analysis is used to analyze alarm data to quickly locate the root cause of the problem and formulate response strategies; and four-quadrant analysis is used to classify the data and clarify its priorities for degree-related data.

[0052] In terms of fault early warning algorithms, linear regression combined with Naive Bayes is mainly used to estimate the probability of system faults occurring within a future period. Collect data related to system failures, including various system performance metrics (such as CPU utilization, memory utilization, network latency, error logs, etc.), and standardize and normalize the data. Use linear regression analysis to analyze the relationship between each feature and system failures. Fit the data using a linear regression model; the general linear regression formula and loss function are shown below:

[0053]

[0054] Obtain the coefficients for each feature and determine the weight of the fault's impact. Based on the results, select the most important features as input and use Naive Bayes for fault classification and prediction. The Naive Bayes formula is as follows:

[0055] Regarding equipment inventory alerts, linear regression and clustering algorithms are used in combination to predict and issue alerts for potential inventory shortages within a certain future timeframe. First, the importance of equipment is categorized using a clustering algorithm; the K-Means clustering is as follows:

[0056] After classifying the devices according to priority, linear regression prediction is performed based on historical information. The formula has been introduced above. Based on the results, a decision tree is used to determine whether an alarm is needed.

[0057] In terms of system security, user authentication and authorization are implemented. To protect system data and functions from unauthorized access, Spring Security is used to implement user authentication and authorization mechanisms. Spring Security provides a robust security framework, supports JWT authentication, and uses BCrypt encryption to store sensitive information, ensuring system security.

[0058] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0059] The foregoing has provided a detailed description of a rail transit fault decision-making system based on an enhanced local AI large model provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are merely for the purpose of helping to understand the method and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application. The above embodiments are merely preferred embodiments given to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention.

Claims

1. A rail transit fault decision-making system based on an enhanced local AI large model, characterized in that, include: The visualization dashboard module provides a user interface for users to interact with backend services, and visualizes and displays fault data and decision analysis information after backend service processing. The data processing and storage module is used to process rail transit fault data and rail transit knowledge, and to build a fault data database and a rail transit knowledge base based on the processed rail transit fault data and rail transit knowledge. The large model building module is used to deploy large AI models locally and create various business workflows and dialogue workflows based on the local large AI models to build various business AI applications, as well as to make fault decisions for various businesses based on rail transit fault data and rail transit knowledge using various business AI applications. The functional interface module provides various business function interfaces to support collaborative work between the data processing and storage module, the large model building module, and the visualization dashboard module, in order to realize various business functions.

2. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The visual dashboard module includes an AI-powered real-time analysis-based decision-making assistant and various types of intelligent chat and dialogue toolsets.

3. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The data processing and storage module is also used to filter the collected rail transit operation and maintenance data to obtain rail transit fault data, preprocess the rail transit fault data, calculate rail transit fault judgment indicators based on the preprocessed rail transit fault data using various rail transit fault judgment index algorithms, and store the rail transit fault data, the preprocessed rail transit fault data, and the rail transit fault judgment indicators into the built database to form a fault data database.

4. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The data processing and storage module is also used to collect rail transit knowledge, perform basic preprocessing and sensitive data filtering on the rail transit knowledge, use segmentation methods to segment and vectorize the processed rail transit knowledge to form a rail transit knowledge base, set up retrieval methods for the rail transit knowledge base, and conduct recall tests on the rail transit knowledge base by analyzing key business needs. Based on the test results, the segmentation and retrieval methods of the rail transit knowledge base are optimized to update and iterate the rail transit knowledge base.

5. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The large model building module constructs business AI applications that use embedded models to vectorize input rail transit fault data and rail transit knowledge. Based on the vectorized input information, it uses the rail transit knowledge base retrieval method to perform similarity retrieval in the rail transit knowledge base. The retrieval results and the vectorized input information are then concatenated in context, and fault decisions are made based on the concatenation results.

6. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The visualization components of the visualization dashboard module should include at least a four-quadrant chart of fault response time, a curve showing the change in lateness rate, the proportion of faulty systems and lines, a heat map of equipment inventory, equipment emergency alarms, various indicator displays, and tables of original fault data.

7. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The visual dashboard module is a visual webpage built using the Vue3 framework with a B / S architecture.

8. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The large model building module is used to build various business AI applications through the locally deployed Dify platform.

9. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The business AI applications built by the large model building module include at least emergency response plan applications, engineering applications, industry standard applications, quality management system applications, system requirement specification applications, business data applications, and code assistants.

10. The rail transit fault decision-making system based on an enhanced local AI large model according to claim 1, characterized in that, The retrieval methods for the rail transit knowledge base include at least hybrid retrieval, which includes vector retrieval and full-text retrieval.

Images