Operation and maintenance decision-making system and method based on large language model and graph database

Abstract

The invention relates to an operation and maintenance decision-making system and method based on a large language model and a graph database, the system comprises an intelligent operation and maintenance platform, a large model layer and an operation and maintenance database, the intelligent operation and maintenance platform is connected with the large model layer and the operation and maintenance database, and the intelligent operation and maintenance platform is used for receiving a user query demand and sending the user query demand to the large model layer. After a task instruction is automatically generated, an operation and maintenance tool is called to execute query analysis operation, and an operation and maintenance processing result is correspondingly returned; the large model layer is used for providing various large model bases so as to correspondingly process task instructions output by the intelligent operation and maintenance platform; and the operation and maintenance database is used for storing the operation and maintenance topological relation data, the operation and maintenance historical event data and the operation and maintenance knowledge data so as to provide a data source for the intelligent operation and maintenance platform. Compared with the prior art, the method has the advantages that the query result can be quickly and accurately obtained, the component fault influence can be determined, the historical operation and maintenance event can be queried and matched, and the operation and maintenance decision efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of operation and maintenance automation technology, and in particular to an operation and maintenance decision-making system and method based on a large language model and a graph database. Background Art

[0002] In the operation and maintenance process of modern enterprises, as the system architecture continues to become more complex, traditional operation and maintenance methods can no longer meet the requirements for fault response speed and decision support. Especially in highly complex industries involving a large number of hosts, applications, components and dependencies, a single operation and maintenance tool is often unable to quickly respond to a variety of operation and maintenance problems.

[0003] In the traditional operation and maintenance management process, only when an event has occurred and has caused business impact can it be discovered and handled. As the complexity of the system increases, operation and maintenance personnel usually need to manually query the component relationships in the system, analyze the scope of the fault impact, and quickly find solutions to similar historical events to assist in decision-making. On the one hand, there is a problem of low efficiency. On the other hand, since the use of graph databases has high technical requirements, it is difficult for many operation and maintenance personnel to manually write Cypher query statements, and it is difficult to effectively conduct in-depth analysis of dependencies and fault impacts in complex systems. These will make it difficult to quickly and accurately obtain query results, evaluate the impact of component failures, and match historical operation and maintenance events in actual applications. It is impossible to obtain reliable reference information, which is not conducive to the efficiency and accuracy of operation and maintenance decisions. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide an operation and maintenance decision-making system and method based on a large language model and a graph database, which can quickly and accurately obtain query results, determine the impact of component failures, and query and match historical operation and maintenance events, thereby improving the efficiency and accuracy of operation and maintenance decisions.

[0005] The object of the present invention can be achieved by the following technical solutions: an operation and maintenance decision system based on a large language model and a graph database, comprising an intelligent operation and maintenance platform, a large model layer and an operation and maintenance database, wherein the intelligent operation and maintenance platform is connected to the large model layer and the operation and maintenance database respectively, and the intelligent operation and maintenance platform is used to receive user query requirements, automatically generate task instructions, call an operation and maintenance tool to perform query analysis operations, and return operation and maintenance processing results accordingly;

[0006] The large model layer is used to provide a variety of large model bases to process the task instructions output by the intelligent operation and maintenance platform accordingly;

[0007] The operation and maintenance database is used to store operation and maintenance topological relationship data, operation and maintenance historical event data and operation and maintenance knowledge data to provide a data source for the intelligent operation and maintenance platform.

[0008] Furthermore, the intelligent operation and maintenance platform is provided with an operation and maintenance intelligent agent, a tool layer and a capability layer. The operation and maintenance intelligent agent is used to generate corresponding task instructions after receiving user query requirements, and autonomously call corresponding operation and maintenance tools in the tool layer;

[0009] The tool layer includes three operation and maintenance tools: graph database query tool, component impact scope tool and operation and maintenance event question and answer tool;

[0010] The capability layer is connected to the big model layer and the operation and maintenance database respectively, and interacts with the data through the big model call to provide support and data processing capabilities for the operation of the tool layer.

[0011] Furthermore, the operation and maintenance database is provided with a graph database, an operation and maintenance event database and an operation and maintenance knowledge base, and the graph database is used to store topological structures and relationship information related to operation and maintenance;

[0012] The operation and maintenance event database is used to store historical operation and maintenance event data;

[0013] The operation and maintenance knowledge base is used to store documents and knowledge related to operation and maintenance.

[0014] Furthermore, the graph database query tool is used to generate and execute a query process corresponding to the user's query requirements, wherein the query process includes clustering query, path query and conditional query;

[0015] The component impact range tool is used to identify and assess the direct and indirect impacts caused by component failure;

[0016] The operation and maintenance event question and answer tool is used to search and match operation and maintenance events, and output similar historical operation and maintenance events and corresponding solutions.

[0017] Furthermore, an embedding model and a vector database are provided in the operation and maintenance event question and answer tool. The embedding model is used to convert the event description of the operation and maintenance event into a vector representation, and store it in the vector database for subsequent retrieval and matching.

[0018] Furthermore, the capability layer is provided with:

[0019] The prompt word engineering module is used to optimize the task description received by the agent, generate query or analysis requirements, and is also used in various operation and maintenance tools to generate prompt words for executing tasks;

[0020] The RAG engine (Retrieval-Augmented Generation) combines retrieval and generation to return specific task execution results based on the operation and maintenance knowledge base and large model;

[0021] Tool calling module, used to coordinate and schedule the calling of various operation and maintenance tools;

[0022] Workflow orchestration module, used to support collaboration between multiple operation and maintenance tools and realize automated task orchestration;

[0023] Multi-model integrated management module, used to manage different large language models and flexibly select and switch large model bases in different tasks;

[0024] The operation and maintenance knowledge base integration module is used to integrate the information in the operation and maintenance knowledge base.

[0025] An operation and maintenance decision method based on a large language model and a graph database includes the following steps:

[0026] Obtain user query requirements, call different operation and maintenance tools, combine the preset big model and operation and maintenance database, and return corresponding visual query results, component failure impact results, similar historical operation and maintenance events and solutions. Among them, the operation and maintenance tools include graph database query tools, component impact range tools and operation and maintenance event Q&A tools.

[0027] Furthermore, the process of returning the visual query result includes:

[0028] According to user query requirements, use the large language model to generate corresponding Cypher query statements;

[0029] Send the generated query statement to the graph database for execution and return the query result;

[0030] After the query results are returned, they are input into the large language model again in combination with the user's query requirements to generate a detailed text response. At the same time, the query results are visualized, the structured data is converted into graphics, and finally the visualization results are displayed to the user together with the text response.

[0031] Furthermore, the process of returning the component failure impact result includes:

[0032] If there are faulty components in the user's query requirements, the prompt word engineering optimization is performed on the large language model to generate optimized Cypher query statements;

[0033] Use the optimized Cypher query statement to query the components affected by the faulty component and their related dependencies from the graph database.

[0034] Furthermore, the process of returning similar historical operation and maintenance events and solutions includes:

[0035] If there are operation and maintenance events in the user's query requirements, each operation and maintenance event is preprocessed, and key information is extracted through a large language model to obtain a refined event description;

[0036] Then, the refined event description is converted into a vector representation through the embedding model and stored in the vectorized database;

[0037] After the user enters the question, the Langchain framework sends the question to the embedding model to convert it into a vector representation. The question vector is sent to the vector database to retrieve the top K historical events that are most similar to it.

[0038] Subsequently, Langchain sends the user’s question and the retrieved Top K events to the large language model, which generates a complete answer and returns the answer to the user through Langchain.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] The present invention constructs an operation and maintenance decision system based on a large language model and a graph database, including an intelligent operation and maintenance platform, a large model layer and an operation and maintenance database. The intelligent operation and maintenance platform is used to receive user query requirements, automatically generate task instructions, call the operation and maintenance tool to perform query analysis operations, and return the operation and maintenance processing results accordingly; the large model layer is used to provide a variety of large model bases to process the task instructions output by the intelligent operation and maintenance platform accordingly; the operation and maintenance database is used to store operation and maintenance topological relationship data, operation and maintenance historical event data and operation and maintenance knowledge data respectively to provide data sources to the intelligent operation and maintenance platform. Thus, by using a graph database and customized operation and maintenance tools, it is possible to assist operation and maintenance personnel to efficiently execute complex queries, analyze the direct and indirect effects of component failures, and quickly match historical operation and maintenance events. Through the natural language processing capabilities of the large language model, the operation and maintenance personnel use daily language to propose query requirements, and the system can automatically generate query statements and execute them, significantly reducing the technical threshold of operation and maintenance work. In addition, the system also supports efficient query and matching of operation and maintenance events, provides accurate reference information, thereby significantly improving operation and maintenance efficiency and meeting the operation and maintenance needs of complex systems.

[0041] The present invention sets up an operation and maintenance intelligent body on the intelligent operation and maintenance platform, which can automatically identify the user's operation and maintenance needs and autonomously select appropriate operation and maintenance tools to perform tasks, thus realizing full-process intelligent operation and maintenance from user questions to automated task completion. Compared with the traditional manual operation and maintenance method, the application of the operation and maintenance intelligent body greatly reduces the technical threshold of operation and maintenance and improves the efficiency of operation and maintenance, especially in cross-component analysis and problem solving of complex systems, which significantly reduces the workload of operation and maintenance personnel.

[0042] The present invention designs a component impact range tool at the tool layer of the intelligent operation and maintenance platform, which can accurately evaluate the impact caused by any component failure in the system based on the optimized query statements generated by the large language model. Through the automatic analysis of complex dependencies in the graph database, the tool realizes a refined evaluation of the impact of component failures, helping operation and maintenance personnel to quickly locate the affected system components and optimize the fault handling process.

[0043] The present invention designs an operation and maintenance event question-and-answer tool at the tool layer of the intelligent operation and maintenance platform. By combining the embedded model and the vector database, the operation and maintenance events input by the user can be vectorized and compared with historical operation and maintenance events, and the most relevant historical events can be quickly retrieved, and specific processing suggestions can be generated. This enables operation and maintenance personnel to refer to solutions of similar historical events when facing new problems, improves the speed and accuracy of decision-making, and effectively reduces the operation and maintenance response time. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of the system framework of the present invention;

[0045] Figure 2 The flowchart of the Chinese database query tool and component impact scope tool of the present invention;

[0046] Figure 3 This is a workflow diagram of the operation and maintenance event question and answer tool in the present invention;

[0047] Figure 4 Schematic diagram of the process of embedding operation and maintenance events into a vector database in the present invention. DETAILED DESCRIPTION

[0048] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example

[0050] This solution proposes an operation and maintenance decision system based on a large language model and a graph database, including an intelligent operation and maintenance platform, a large model layer, and an operation and maintenance database. The intelligent operation and maintenance platform is connected to the large model layer and the operation and maintenance database respectively. The intelligent operation and maintenance platform is used to receive user query requirements, automatically generate task instructions, call the operation and maintenance tool to perform query analysis operations, and return the operation and maintenance processing results accordingly;

[0051] The large model layer is used to provide a variety of large model bases to process the task instructions output by the intelligent operation and maintenance platform;

[0052] The operation and maintenance database is used to store operation and maintenance topology relationship data, operation and maintenance historical event data, and operation and maintenance knowledge data to provide a data source for the intelligent operation and maintenance platform.

[0053] The above system is applied in practice to realize an operation and maintenance decision-making method based on a large language model and a graph database, which includes the following steps: obtaining user query requirements, calling different operation and maintenance tools, combining the preset large model and the operation and maintenance database, and correspondingly returning visual query results, component failure impact results, similar historical operation and maintenance events and solutions. Among them, the operation and maintenance tools include graph database query tools, component impact range tools and operation and maintenance event question and answer tools.

[0054] This embodiment applies the above solution to build Figure 1 The application framework shown mainly includes:

[0055] 1. Operation and maintenance agent

[0056] The operation and maintenance agent is the core module of the entire system in this solution. It is responsible for scheduling and managing the operation of various tools and providing intelligent operation and maintenance services. The operation and maintenance agent can autonomously call different tools, identify the intention based on the user's questions, and call the appropriate tool module for query and analysis, and finally return the results. This module receives user input or operation and maintenance events, calls the graph database query tool, component impact range tool or operation and maintenance event question and answer tool, and returns the processing results.

[0057] It should be explained that this solution uses agents and tools based on large language models. Among them, agents based on large language models are intelligent agents that can perform tasks autonomously and complete complex operation and maintenance tasks by calling different tool modules. In this solution, the agent identifies the user's intention and selects appropriate tools (such as database query tools, component impact range tools, and operation and maintenance event question and answer tools) to perform tasks, thereby realizing automated operation and maintenance from problem raising to final resolution. The operation and maintenance tool is a specific operation and maintenance execution unit. Each tool is designed for different operation and maintenance needs. By cooperating with the large language model, the intelligence and automation level of operation and maintenance work are improved.

[0058] 2. Tool layer

[0059] The tool layer includes three main operation and maintenance tools, each corresponding to different operation and maintenance functions: graph database query tool, component impact scope tool, and operation and maintenance event Q&A tool. The workflow of the graph database query tool and component impact scope tool is as follows: Figure 2 As shown in the figure, the workflow of the operation and maintenance event question and answer tool is as follows Figure 3 shown.

[0060] The graph database query tool can generate the corresponding query process based on the questions entered by the user. First, the user puts forward the query requirements, and the large language model (LLM) in the tool generates the appropriate Cypher query statement. The generated query statement is sent to the graph database (such as Neo4j) for execution, and the query results are returned. After the query results are returned, they are input into the large language model again in combination with the user's questions to generate a detailed text response. At the same time, the tool visualizes the query results, converts structured data into easy-to-understand graphics, and finally presents the visualization results to the user together with the text response. The tool can efficiently execute various types of aggregate queries (for example, counting the number of applications in the system, or the number of hosts in each network), path queries (for example, calculating the shortest path from Server A to App A), and conditional queries (for example, finding all running databases and their dependent databases).

[0061] It should be explained that the graph database (Neo4j) is a database that stores data in a graph structure, which is used to represent and query entities and the complex relationships between them. In this solution, Neo4j is used to store operation and maintenance data such as hosts, applications, components, configurations, etc. in the system, and the Cypher query language is used to achieve efficient retrieval and analysis of data. Cypher query language is a query language used by Neo4j, similar to SQL, and is used to query nodes and relationships in the graph database. In this solution, Cypher is used to generate complex query statements to help operation and maintenance personnel extract the required information from the graph database. The Large Language Model (LLM) is a natural language processing model based on deep learning that can understand and generate natural language text. In this solution, the large language model is used to convert the user's natural language input into structured query instructions (such as Cypher statements) and generate detailed answers to questions to lower the technical threshold of operation and maintenance personnel.

[0062] The component impact range tool is specifically used to identify and evaluate the direct and indirect impact of component failures on the system. During the operation and maintenance process, when a component fails, it is usually necessary to quickly understand the scope of its impact on other system components. When dealing with such relatively vague queries, the query statements generated by general graph database query tools often lack sufficient accuracy. In order to solve this high-frequency usage scenario, this solution has specially developed a component impact range tool to optimize the prompt word engineering of the large language model in order to generate more accurate Cypher query statements. The tool can identify the direct and indirect impacts of component failures, query the affected components and their related dependencies, and help the operation and maintenance team quickly locate problems and evaluate their potential impacts, so as to formulate effective countermeasures.

[0063] It should be explained that prompt engineering is a method to optimize the output of large language models. By designing and adjusting input prompts, the model's response to specific tasks is improved. In this solution, prompt engineering is used to generate more accurate query statements to help the tool layer complete operation and maintenance tasks efficiently.

[0064] The operation and maintenance event question and answer tool combines the embedded model with the vector database to quickly provide users with analysis of relevant operation and maintenance events. Figure 4 As shown in the figure, the tool first preprocesses each operation and maintenance event. The event is input in the form of original description. After the key information is extracted by the large language model, a refined event description is obtained. Then, the refined event description is converted into a vector representation through the embedding model and stored in the vectorized database. This process enables each event to be stored in the form of a vector for subsequent rapid retrieval and matching. After the user enters the question, the Langchain framework sends the question to the embedding model and converts it into a vector representation so that similar events can be retrieved in the vectorized database. Then, the question vector is sent to the vector database to retrieve the Top K historical events that are most similar to it. Subsequently, Langchain sends the user's question and the retrieved Top K events to the large language model, and the large language model generates a complete answer. Finally, the answer is returned to the user through Langchain, helping the user to quickly find operation and maintenance events similar to the current problem and their solutions. The tool can efficiently handle similar event matching tasks, provide accurate reference information, and effectively improve operation and maintenance efficiency.

[0065] It should be explained that the embedding model is a method of converting text information into a vector representation so that the text can be understood and processed by the machine learning model. In this solution, the embedding model is used to convert the description of the operation and maintenance event into a vector form in order to retrieve similar events in the vector database. The Langchain framework is a framework for building and managing language model applications, which can integrate multiple language models and tools to automate complex tasks. In this solution, the Langchain framework is used to coordinate the interaction between large language models, embedding models, and vector databases to complete the query and answer of operation and maintenance events. The vector database is a database for storing and retrieving vectorized data, which is suitable for scenarios such as similarity search. In this solution, the vector database is used to store embedded operation and maintenance events so that similar events can be quickly retrieved when users query.

[0066] Each tool in the tool layer processes user requests or events, generates corresponding queries or analysis tasks, and interacts with the data layer to achieve specific functional requirements. Each tool module receives task instructions through close collaboration with the capability layer and the data layer, extracts or analyzes the required information from relevant data sources, and finally returns the processing results to the operation and maintenance agent to complete the overall task.

[0067] 3. Capability layer

[0068] The capability layer is responsible for providing various capability support to the tool layer and improving the intelligence level of the system. Prompt engineering is not only used to optimize the task description received by the agent and generate more accurate query or analysis requirements, but is also used in various tools to generate prompts for executing tasks to improve the tool's response effect and task completion. The RAG engine combines retrieval and generation, and returns specific task execution results based on the operation and maintenance knowledge base and the large model. The RAG engine is a technology that combines retrieval and generation. It retrieves relevant information from the knowledge base and generates the final answer in combination with the large language model. In this solution, the RAG engine is used to retrieve relevant information from the operation and maintenance knowledge base and the graph database, and generates accurate query results and answers in combination with the large language model, thereby improving the efficiency of information acquisition in the operation and maintenance process; the tool call module is responsible for coordinating and scheduling the calls of various tool modules so that tasks can be executed efficiently; workflow orchestration supports collaboration between multiple tools, and optimizes workflow execution efficiency through task automation orchestration; the multi-model integration management module is used to manage different large language models, and supports the platform to flexibly select and switch large model bases in different tasks; the operation and maintenance knowledge base integration module integrates information in the operation and maintenance knowledge base to support problem analysis and troubleshooting. The capability layer interacts with data through large model calls, providing support and data processing capabilities for the operation of the tool layer.

[0069] 4. Large model layer

[0070] The large model layer provides the system with a foundation for natural language processing and task generation, and completes complex task generation and analysis by calling large language models. The large model layer receives the task description generated by the capability layer, and generates corresponding query instructions or operation and maintenance suggestions through natural language understanding to support the execution of various tasks of the system. This layer provides different large model bases, such as ChatGLM, Baichuan, and Qwen, for processing task requirements.

[0071] 5. Data Layer

[0072] The data layer is the core data source of the system, providing structured and unstructured data support required for operation and maintenance operations. The graph database (Neo4j) is used to store and manage topological structures and relationship information related to operation and maintenance, including key operation and maintenance data such as hosts, applications, components, and configurations, and supports query operations of graph database query tools and component impact range tools; the operation and maintenance event database stores historical operation and maintenance events for similarity analysis and embedded matching by the operation and maintenance event question and answer tool; the operation and maintenance knowledge base stores documents and knowledge related to operation and maintenance, and supports RAG engine retrieval and problem analysis. The data layer supports the operation of various tools and modules by providing efficient and structured operation and maintenance data. The graph database provides topological relationship data, the operation and maintenance event database provides historical events, and the operation and maintenance knowledge base provides operation and maintenance experience and documents. All data is processed and queried through the RAG engine and tool layer.

[0073] In summary, the intelligent operation and maintenance platform in this solution is jointly constructed by the operation and maintenance intelligent body, tool layer, capability layer, large model layer and data layer. Each module works closely together to form an efficient and intelligent operation and maintenance ecosystem. Each module has specific responsibilities in the system. Through mutual cooperation, the whole process of automated operation and maintenance from problem raising to final solution is realized, which significantly improves the efficiency and intelligence level of operation and maintenance.

Claims

1. An operation and maintenance decision system based on a large language model and a graph database, characterized in that: It includes an intelligent operation and maintenance platform, a large model layer and an operation and maintenance database. The intelligent operation and maintenance platform is connected to the large model layer and the operation and maintenance database respectively. The intelligent operation and maintenance platform is used to receive user query requirements, automatically generate task instructions, call the operation and maintenance tool to perform query analysis operations, and return the operation and maintenance processing results accordingly; The large model layer is used to provide a variety of large model bases to process the task instructions output by the intelligent operation and maintenance platform accordingly; The operation and maintenance database is used to store operation and maintenance topological relationship data, operation and maintenance historical event data and operation and maintenance knowledge data to provide a data source for the intelligent operation and maintenance platform.

2. According to claim 1, an operation and maintenance decision system based on a large language model and a graph database is characterized in that: The intelligent operation and maintenance platform is provided with an operation and maintenance intelligent agent, a tool layer and a capability layer. The operation and maintenance intelligent agent is used to generate corresponding task instructions after receiving user query requirements, and autonomously call corresponding operation and maintenance tools in the tool layer; The tool layer includes three operation and maintenance tools: graph database query tool, component impact scope tool and operation and maintenance event question and answer tool; The capability layer is connected to the big model layer and the operation and maintenance database respectively, and interacts with the data through the big model call to provide support and data processing capabilities for the operation of the tool layer.

3. According to claim 2, an operation and maintenance decision system based on a large language model and a graph database is characterized in that: The operation and maintenance database is provided with a graph database, an operation and maintenance event database and an operation and maintenance knowledge base, and the graph database is used to store topological structures and relationship information related to operation and maintenance; The operation and maintenance event database is used to store historical operation and maintenance event data; The operation and maintenance knowledge base is used to store documents and knowledge related to operation and maintenance.

4. According to claim 3, an operation and maintenance decision system based on a large language model and a graph database is characterized in that: The graph database query tool is used to generate and execute a query process corresponding to the user's query requirements, and the query process includes clustering query, path query and conditional query; The component impact range tool is used to identify and assess the direct and indirect impacts caused by component failure; The operation and maintenance event question and answer tool is used to search and match operation and maintenance events, and output similar historical operation and maintenance events and corresponding solutions.

5. According to claim 4, an operation and maintenance decision system based on a large language model and a graph database is characterized in that: The operation and maintenance event question and answer tool is provided with an embedded model and a vector database. The embedded model is used to convert the event description of the operation and maintenance event into a vector representation and store it in the vector database for subsequent retrieval and matching.

6. According to claim 3, the operation and maintenance decision system based on a large language model and a graph database is characterized in that: The capability layer is provided with: The prompt word engineering module is used to optimize the task description received by the agent, generate query or analysis requirements, and is also used in various operation and maintenance tools to generate prompt words for executing tasks; The RAG engine combines retrieval and generation to return specific task execution results based on the operation and maintenance knowledge base and large model; Tool calling module, used to coordinate and schedule the calling of various operation and maintenance tools; Workflow orchestration module, used to support collaboration between multiple operation and maintenance tools and realize automated task orchestration; Multi-model integrated management module, used to manage different large language models and flexibly select and switch large model bases in different tasks; The operation and maintenance knowledge base integration module is used to integrate the information in the operation and maintenance knowledge base.

7. An operation and maintenance decision method based on a large language model and a graph database, applied to an operation and maintenance decision system based on a large language model and a graph database as claimed in claim 1, characterized in that: The following steps are involved: Obtain user query requirements, call different operation and maintenance tools, combine the preset big model and operation and maintenance database, and return corresponding visual query results, component failure impact results, similar historical operation and maintenance events and solutions. Among them, the operation and maintenance tools include graph database query tools, component impact range tools and operation and maintenance event Q&A tools.

8. The operation and maintenance decision method based on a large language model and a graph database according to claim 7, characterized in that: The process of returning the visual query result includes: According to user query requirements, use the large language model to generate corresponding Cypher query statements; Send the generated query statement to the graph database for execution and return the query result; After the query results are returned, they are input into the large language model again in combination with the user's query requirements to generate a detailed text response. At the same time, the query results are visualized, the structured data is converted into graphics, and finally the visualization results are displayed to the user together with the text response.

9. The operation and maintenance decision method based on a large language model and a graph database according to claim 7, characterized in that: The process of returning the component failure impact result includes: If there are faulty components in the user's query requirements, the prompt word engineering optimization is performed on the large language model to generate optimized Cypher query statements; Use the optimized Cypher query statement to query the components affected by the faulty component and their related dependencies from the graph database.

10. The operation and maintenance decision method based on a large language model and a graph database according to claim 7, characterized in that: The process of returning similar historical operation and maintenance events and solutions includes: If there are operation and maintenance events in the user's query requirements, each operation and maintenance event is preprocessed, and key information is extracted through a large language model to obtain a refined event description; Then, the refined event description is converted into a vector representation through the embedding model and stored in the vectorized database; After the user enters the question, the Langchain framework sends the question to the embedding model to convert it into a vector representation. The question vector is sent to the vector database to retrieve the top K historical events that are most similar to it. Subsequently, Langchain sends the user’s question and the retrieved Top K events to the large language model, which generates a complete answer and returns the answer to the user through Langchain.

Images