Multi-agent-based cross-station intelligent inspection and operation and maintenance method and system

Abstract

The invention discloses a multi-agent-based cross-station intelligent inspection and operation and maintenance method and system, and the system comprises an inspection expert agent, a tool use agent, a code generation agent, an overhaul expert agent and a memory module. Calling a tool using agent to obtain corresponding power station information generation suggestions, and issuing corresponding tasks after user confirmation; a code generation agent receives the task, a tool use agent is called to obtain a corresponding power station model version and hardware equipment information generation script, and script parameters are dynamically adjusted; and the maintenance expert agent obtains the script operation detection model, calls the tool use agent to obtain corresponding power station monitoring information, detects the monitoring information by using the detection model, stores the detection result to the short-term memory module, and gives a maintenance suggestion and adjusts a detection strategy by combining the detection result with the maintenance knowledge information in the long-term memory module. According to the invention, efficient, intelligent and sustainable automatic task management and execution are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to artificial intelligence technology, in particular to a multi-agent based cross-station intelligent inspection and operation method and system. BACKGROUND

[0002] With the continuous improvement of China's comprehensive national strength and technological level, the dependence of modern industrial system and residents' life on electricity is increasing. Whether in high-energy manufacturing, data centers and other key industries, or in smart home, transportation, communication and other daily fields, higher requirements are put forward for the safe, stable and efficient supply of electric energy. To meet this trend, the power grid system has accelerated the pace of infrastructure construction, especially the layout density and service radius of the substation have put forward higher coverage requirements, resulting in the continuous growth of the number of substations, the more extensive spatial distribution, and the increasingly complex and variable operating environment. Under this background, how to ensure the long-term stable operation of various substation facilities has become the core goal of power enterprise operation and maintenance. However, the current operation and maintenance of substations still relies heavily on manual regular inspection, on-site observation and manual configuration, which is not only low in efficiency and long in cycle, but also has problems such as untimely response, unreasonable resource allocation, weak fault warning capability, etc. Especially in the large-scale power grid system with multiple sites running in parallel, the traditional way cannot meet the efficient collaborative operation and maintenance demand in the cross-regional and heterogeneous device environment. At present, although some automatic inspection systems have been put into use, they are mainly based on "preset script + rule trigger", which cannot effectively understand the semantics of operation and maintenance tasks, lack flexibility and dynamic adjustment capability, and still rely on manual intervention and rule maintenance. More importantly, most of the traditional systems are designed for single-site deployment, lacking a unified cross-site scheduling framework, resulting in high deployment and maintenance cost, low task distribution efficiency, and serious data island, which cannot realize knowledge sharing, experience reuse and global intelligent optimization among sites. In addition, the inspection task planning method is single, most systems do not have the ability to dynamically adjust strategies according to environmental, historical data, real-time state and other multi-dimensional information, and the operation and maintenance process lacks a self-adaptive closed-loop mechanism, making it difficult to form an intelligent agent system with continuous learning and evolution capability. Finally, the traditional inspection and operation and maintenance system not only cannot guarantee the operation and maintenance quality, but also causes a lot of waste of human resources and system response lag, which cannot meet the core requirements of "high reliability, high efficiency and strong collaboration" of power operation and maintenance in the new era. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a multi-agent based cross-station intelligent inspection and operation method and system to realize adaptive operation and maintenance and closed-loop optimization in a variable environment, and comprehensively improve the intelligent, collaborative and efficient level of power system operation and maintenance.

[0004] To solve the above technical problems, the technical scheme adopted by the present application is: A multi-agent-based cross-site intelligent inspection and maintenance method is proposed. The method is applied to a cross-site intelligent inspection and maintenance system comprising an inspection expert agent, a tool-using agent, a code-generating agent, a maintenance expert agent, and a memory module. The method includes the following steps: The inspection expert intelligent agent receives the inspection and maintenance requirements input by the user, performs semantic parsing through a large language model, extracts structured information, then calls the tool to use the intelligent agent to obtain the power station information that matches the structured information content and generates inspection and maintenance suggestions. Finally, the corresponding power station equipment status and inspection and maintenance suggestions are displayed through the interactive page, and the corresponding inspection and maintenance tasks are issued after the user confirms. The code generation agent receives inspection and maintenance tasks, calls tools to use the agent to obtain the model version information and hardware equipment information of the corresponding power station, and then, based on the model version information and hardware equipment information, retrieves the corresponding model configuration and version from the long-term memory module of the memory module to generate an execution script, and dynamically adjusts the script parameters according to the information of the inspection and maintenance tasks. The maintenance expert agent obtains the execution script, initializes and runs the detection model of the power plant site corresponding to the inspection and maintenance task according to the execution script information, calls the tool to use the agent to obtain the monitoring information of the corresponding power plant site, uses the detection model to detect the monitoring information, and saves the detection results to the short-term memory module of the memory module. The maintenance expert agent provides maintenance suggestions and adjusts the detection strategy based on the detection results in the short-term memory module and the maintenance knowledge information stored in the long-term memory module.

[0005] Furthermore, the structured information includes time information, location information, and weather information. When the inspection expert agent invokes tools to obtain power station information matching the structured information content and generates inspection and maintenance suggestions, it specifically includes: The inspection expert intelligent agent uses the time and location information as the target time and target location, and calls the tool to use the intelligent agent to obtain the equipment information and weather information of all power stations at the target time and target location; The tool uses an intelligent agent to attempt to call the equipment of all power stations at the target time and target location to obtain the equipment information of available equipment. It returns the equipment information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The tool also uses an intelligent agent to call the field sensors of all power stations at the target time and target location to obtain weather information. It returns the equipment availability information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The inspection expert intelligent agent matches the weather information in the structured information with the weather information of all power stations at the target time and target location stored in the short-term memory module to obtain the location and equipment information of the substation that matches the current weather conditions; The inspection expert agent combines the substation's time and location information, which are consistent with the current weather conditions, with the inspection and maintenance requirements input by the user. It matches the corresponding information in the long-term memory module. Based on the matching results and system prompts, it uses a language model to generate inspection suggestions. The inspection suggestions include key areas and equipment to focus on. Then, by combining the inspection suggestions with the time, location, and weather information in the structured information, and the equipment information stored in the short-term memory module, it attempts to obtain the key equipment information to focus on under the current weather conditions.

[0006] Furthermore, the structured information includes time information and location information. When the inspection expert agent calls the tool to obtain power station information matching the structured information content and generates inspection and maintenance suggestions, it specifically includes: The inspection expert intelligent agent uses the time and location information as the target time and target location, and calls the tool to use the intelligent agent to obtain the equipment information and weather information of all power stations at the target time and target location; The tool uses an intelligent agent to attempt to call the equipment of all power stations at the target time and target location to obtain the equipment information of available equipment. It returns the equipment information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The tool also uses an intelligent agent to call the field sensors of all power stations at the target time and target location to obtain weather information. It returns the equipment availability information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The inspection expert intelligent agent combines weather information with the user's inspection and maintenance needs, matches the corresponding information in the long-term memory module, and provides inspection suggestions using a language model based on the matching results and system prompts. The inspection suggestions include key areas and equipment to focus on. Then, by combining the inspection suggestions with the previously obtained location and time information, as well as the equipment information of each power station stored in the short-term memory module, it attempts to obtain the key equipment information of the power station at the current location.

[0007] Furthermore, the structured information includes information related to the comprehensive inspection. When the inspection expert agent invokes tools to obtain power plant information matching the structured information content and generates inspection and maintenance suggestions, it specifically includes: The inspection expert agent calculates the current time and calls upon tools to obtain equipment information for all power plants at the current time. The tool uses an intelligent agent to attempt to call the equipment of all power stations at the current time, obtain the equipment information of available equipment, return the equipment information of each power station to the inspection expert intelligent agent and save it to the short-term memory module; The inspection expert intelligent agent combines the user's inspection and maintenance needs with the corresponding information in the long-term memory module. Based on the matching results, it uses a language model to summarize the inspection content, including the inspection equipment, inspection time, inspection location, inspection priority, inspection type, and execution method, along with system prompts.

[0008] Furthermore, the structured information includes information unrelated to inspection and maintenance work. When the inspection expert agent calls the tool to obtain power plant information matching the structured information content and generates inspection and maintenance suggestions, the specific steps include: If the structured information contains information unrelated to inspection and maintenance work, the inspection expert intelligent agent will prompt the user to re-enter the requirements through an interactive page.

[0009] Furthermore, after the code generation agent obtains the corresponding model configuration and version from the long-term memory module of the memory module to generate the execution script, it also saves the corresponding script template. After dynamically adjusting the script parameters according to the information of the inspection and maintenance tasks, it also includes: the code generation agent applying a large language model to find common interfaces and common parameters for the generated execution script templates, and optimizing and unifying them to obtain a unified template representation.

[0010] Furthermore, when the maintenance expert agent provides maintenance suggestions and adjusts the testing strategy based on the test results in the short-term memory module and the maintenance knowledge information stored in the long-term memory module, it specifically includes: The maintenance expert agent uses a language model to generate content for adjusting the training strategy based on the maintenance knowledge information stored in the long-term memory module and the test results in the short-term memory module, according to a specified period. It also reads the latest strategy content and modifies the test strategy accordingly. The maintenance expert agent queries all relevant information in the short-term memory module and applies a large language model, combined with maintenance knowledge information stored in the long-term memory module, to provide comprehensive maintenance suggestions.

[0011] Furthermore, after saving the test results to the short-term memory module of the memory module, the process also includes: after the test is completed, the maintenance expert agent calls the large language model to summarize the test results in the short-term memory module and transfers them to the long-term memory module, thereby updating the maintenance knowledge information stored in the long-term memory module for reference in subsequent test work.

[0012] Furthermore, after the maintenance expert agent provides maintenance suggestions and adjusts the detection strategy based on the detection results in the short-term memory module and the maintenance knowledge information stored in the long-term memory module, the process also includes: the maintenance expert agent feeding back the maintenance suggestions to the inspection expert agent, which then displays the maintenance suggestions through an interactive page.

[0013] This invention also proposes a multi-agent-based cross-site intelligent inspection and maintenance system, including an inspection expert agent, a tool-using agent, a code-generating agent, a maintenance expert agent, and a memory module. The multi-agent-based cross-site intelligent inspection and maintenance system is programmed or configured to execute the steps of the multi-agent-based cross-site intelligent inspection and maintenance method.

[0014] Compared with the prior art, the advantages of the present invention are as follows: This invention achieves fully closed-loop automated processing of natural language task parsing, equipment scheduling, data acquisition, fault detection, and intelligent maintenance suggestions, wherein: In this invention, the inspection expert agent is responsible for demand analysis and inspection suggestions, the code generation agent dynamically generates scripts based on the tasks, and the maintenance expert agent executes the inspections and provides feedback for adjustments. Each agent has a clear division of labor and cooperates with the others to form an efficient operation and maintenance process. The code generation agent of this invention dynamically adjusts script parameters based on the power plant model and hardware information, while the maintenance expert agent combines real-time monitoring information and long-term maintenance knowledge to dynamically optimize detection strategies and maintenance suggestions, enabling the system to quickly adapt to environmental changes. The short-term memory module of this invention stores real-time detection results, while the long-term memory module accumulates historical maintenance experience. When processing tasks, the maintenance expert agent refers to the knowledge in the long-term memory and combines it with new information in the short-term memory to continuously update and optimize knowledge, thereby improving the accuracy and efficiency of operation and maintenance decisions. Attached Figure Description

[0015] Figure 1 This is a block diagram of the system structure in an embodiment of the present invention.

[0016] Figure 2 This is a simplified flowchart of the method in an embodiment of the present invention.

[0017] Figure 3 This is a detailed flowchart of the method in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and specific preferred embodiments, but this does not limit the scope of protection of the present invention.

[0019] To address the shortcomings of existing methods, such as high labor costs and low efficiency, time-consuming and labor-intensive processes, independent site-level deployments, lack of information sharing between different sites, inability to achieve unified planning, resource scheduling, and task collaboration across sites, and weak ability to handle emergencies and lack of intelligent capabilities, this embodiment proposes a multi-agent-based cross-site intelligent inspection and maintenance method. Figure 1The system shown is a cross-site intelligent inspection and maintenance system that includes an inspection expert agent, a tool-using agent, a code-generating agent, a maintenance expert agent, and a memory module. Based on a multi-agent architecture, it supports functions such as natural language task input, automatic task planning, equipment adaptation and execution, intelligent fault judgment, and experience summary and transfer. It enables adaptive operation and maintenance and closed-loop optimization in a variable environment, and comprehensively improves the intelligence, collaboration, and efficiency of power system operation and maintenance.

[0020] Figure 1 The system shown is a novel inspection and maintenance system with cross-site scheduling capabilities, intelligent task understanding and automated execution capabilities, and autonomous knowledge learning and evolution capabilities. It consists of four intelligent agents: an inspection expert agent, a maintenance expert agent, a tool usage agent, and a code generation agent. It can achieve closed-loop automated processing of the entire process, from natural language task parsing, equipment scheduling, data acquisition, fault detection, and intelligent maintenance suggestions. The system incorporates a memory block, including a short-term memory (STM) module and a long-term memory (LTM) module, to support experience accumulation and knowledge reuse. Specifically: Inspection Expert Agent: Primarily responsible for task understanding and intelligent planning. As a core intelligent agent in this system, it is responsible for interacting with users, providing result feedback, and coordinating the work of various intelligent agents. Its main functions include receiving natural language descriptions input by users, performing semantic parsing through a large language model, extracting structured information (including time, location, and whether weather requirements are included), and scheduling the Long Term Memory (LTM) module within the Memory module to query historical experience, obtain precautions, risk warnings, key equipment and their priorities for the current scenario, and dynamically adjust all information based on the results obtained from the Memory module or subsequent user input. Therefore, it is also allowed to interact with users, modify inspection and maintenance plans, and ultimately determine the inspection and maintenance schedule, coordinating the execution of downstream agents.

[0021] Maintenance Expert Agent: Performs inspections and returns diagnostic suggestions. Its main functions include receiving task plans from superior agents, executing device inspections based on image / sensor data, analyzing and identifying the results, determining device status, providing maintenance suggestions, and writing inspection results in real-time to the Short-Term Memory (STM) module of the Memory module for strategy optimization in the current task phase. Finally, after the inspection and maintenance are completed, key features in the STM are summarized, and this information is updated in the Long-Term Memory (LTM).

[0022] Tool Usage Agent: Tool Invocation and Data Acquisition. Its main function is to invoke field equipment (such as cameras, drones, environmental monitors, etc.) for data collection. Based on the needs of the inspection task, it can determine weather conditions before inspection and maintenance, identify and dispatch suitable physical equipment, and return equipment status and weather information to the inspection agent. Simultaneously, it collects image or sensor data in real time during the task and returns the results to the maintenance expert agent. It supports cross-site equipment scheduling and provides a unified hardware invocation interface for multiple sites.

[0023] Code Generation Agent: Automatically generates detection execution scripts. Based on the task plan (location, device type, priority), it can obtain the corresponding model configuration and version from the Long Memory Module (LTM) to dynamically generate an execution script adapted to the current task. At the same time, it accumulates the script templates used during long-term operation and optimizes the script generation logic.

[0024] Memory modules: including Short-Term Memory (STM) and Long-Term Memory (LTM): 1. STM records real-time data and feedback during the current task execution, including the number of anomalies identified, information on the identified objects, weather conditions, substation information, etc.

[0025] 2. LTM summarizes the results of multiple rounds of tasks, automatically summarizes the experience rules through a large language model, and updates them to the knowledge base for future task scheduling reference.

[0026] The aforementioned agents work together to form a cross-agent scheduling system. The system has an inter-agent collaboration protocol stack, supports information sharing, instruction issuance and task feedback mechanisms, and can realize a closed-loop collaborative process of task planning, script generation, device scheduling, detection execution and data summarization.

[0027] like Figure 2 As shown, based on Figure 1 The system in this embodiment includes the following steps: S1) Obtain user inspection and maintenance requirements. Specifically, this involves guiding users to input their desired inspection and maintenance tasks through an interactive page. S2) Analyze the time and location of the inspection and maintenance requirements, query and filter the equipment information of the substation to be inspected, and provide relevant inspection suggestions. Users can adjust their inspection needs according to the suggestions. Specifically, the inspection and maintenance requirements are input into the Inspection Expert Agent. The Inspection Expert Agent receives the inspection and maintenance requirements input by the user, performs semantic parsing of the input text using a large language model, extracts structured information, and then calls the tool to obtain information about the substation that matches the structured information content and generates inspection and maintenance suggestions. Finally, the corresponding substation's equipment status and inspection and maintenance suggestions are displayed through an interactive page. Users can change their inspection and maintenance requirements on this interactive page. After the user confirms the inspection and maintenance task, the corresponding inspection and maintenance task will be issued. S3) Query information such as on-site tools and models at the inspection and maintenance location, and generate inspection scripts based on the video, image, and other information returned by the on-site tools. Specifically, the code generation agent receives the inspection and maintenance task and calls the tool to use the agent to obtain the model version information and hardware equipment information of the corresponding power station. Based on the inspection and maintenance task information, the tool uses the agent to query the model version information deployed at the relevant site and the on-site hardware equipment information, and obtains video or images from the on-site hardware equipment such as cameras. At the same time, the obtained model version information is passed to the code generation agent. Then, the code generation agent obtains the corresponding model configuration and version from the LTM of the memory module based on the model version information and hardware equipment information to generate the execution script, and dynamically adjusts the script parameters according to the inspection and maintenance task information. S4) Execute inspection tasks and dynamically adjust inspection strategies during the process, ultimately returning inspection results and suggestions while updating the memory module. Specifically, the execution script is passed to the maintenance expert agent for inspection work. After receiving the execution script, the maintenance expert agent selects different power plant detection models based on the script content, performs inspections on specific field equipment, returns the inspection results, inputs them into the STM, dynamically adjusts the detection strategy based on the current output results, and returns maintenance suggestions after the entire inspection is completed. It also summarizes the information in the STM using a large language model and inputs it into the LTM for storage. Specifically, it initializes and runs the detection model corresponding to the power plant site for the inspection and maintenance tasks based on the execution script information, calls the tool to use the agent to obtain the monitoring information of the corresponding power plant site, uses the detection model to detect the monitoring information, and saves the detection results to the STM in the memory module. The maintenance expert agent, based on the detection results in the STM and the maintenance knowledge information stored in the LTM, comprehensively provides maintenance suggestions and adjusts the detection strategy.

[0028] The following is in conjunction with the appendix Figure 3 The document provides a detailed explanation of the functions of different intelligent agents and their workflows in each step.

[0029] In this embodiment, the main function of the inspection expert agent is to receive natural language descriptions input by the user, perform semantic parsing through a large language model, extract structured information, schedule various intelligent agents to work together, and can also link with the long-term memory module in the Memory module to query historical experience, obtain precautions, risk warnings, key equipment and their priorities in the current scenario, and dynamically adjust all information according to the results obtained by the Memory module or the user's subsequent input requirements. Therefore, it is also allowed to interact with the user, modify the inspection and maintenance plan, and finally determine the inspection and maintenance plan and coordinate the execution of downstream agents.

[0030] In this embodiment, the inspection and maintenance task includes inspection area information, task time plan, list of target equipment, and environmental condition requirements. The inspection area information is used to limit the substations or multiple cross-site inspection scenarios applicable to this task. The environmental conditions include external conditions such as weather, time period, and load status, used to dynamically adapt appropriate detection models and rule templates. The inspection task configuration rules include a historical experience rule base generated based on long-term memory (LTM), which includes the following: 1. Potential fault scenarios under different weather conditions, along with recommendations and knowledge base related to inspection and maintenance; 2. Inspection and maintenance strategies and optimization records for historical patrol tasks; 3. On-site information such as equipment model versions for each power station in the system; Therefore, in step S2, the inspection expert Agent, based on the user's natural language input of scenario requirements, geographical location, and time, completes semantic parsing through a large language model, and performs device screening, rule matching, and task structured configuration based on the LTM memory content, outputting a complete inspection and maintenance suggestion plan. Its main execution flow is as follows: 1. Receive user inspection and maintenance requests It receives text input from the user and uses a large language model to process the input text information and determine the user's content, specifically categorized as follows: (1) The information related to substation inspection data in this question is related to the knowledge of substation inspection safety data and the call strategy for inspection in abnormal weather. That is, the input information does not include weather conditions.

[0031] (2) This problem is not related to the knowledge of substation inspection safety data, nor to the abnormal weather inspection call strategy, and may be due to incorrect input.

[0032] (3) The question states the weather conditions, which is related to the abnormal weather inspection call strategy. That is, the question contains weather information, and the non-comprehensive inspection work does not need to be carried out for all power plants.

[0033] (4) Comprehensive inspection: There may be leaders inspecting the area. A comprehensive inspection of the designated area is conducted without paying attention to the weather. Just return to the equipment that needs to be inspected.

[0034] Subsequently, different processing methods were applied to the four different needs mentioned above. The following sections demonstrate the different follow-up processing methods for the four needs.

[0035] 2. Address different needs. (1) Processing flow for requests that do not include weather information The location and time information are extracted using a large language model. Note that the time information is always in the YYYY-MM-DD format.

[0036] The inspection expert agent coordinates and calls the tool to use the agent, inputting the location, city and time information obtained here, to obtain the equipment information and weather information of all substations in the area, and returns the information to the inspection expert agent, while saving this information to the short-term memory (STM) of the memory module.

[0037] The inspection expert agent will combine the current weather conditions with the inspection and maintenance requirements input by the user, use a language model to organize the complete inspection and maintenance requirements, and combine the long-term memory (LTM) of the memory module to match relevant information.

[0038] Based on the matching results and system prompts, the system provides inspection suggestions to the user using a language model. The system prompt is: "You are an expert in the field of substation inspection safety knowledge in the company. The user asked about which areas and equipment to pay attention to during severe inspections, as well as inspection safety knowledge. You answer based on the knowledge queried below." The inspection suggestions include key areas and equipment to focus on.

[0039] Based on the inspection recommendations, the previously obtained location and time information, and the equipment information of each power station stored in the Short-Term Memory (STM), the system attempts to obtain the key equipment information of the power station at the current location. If the system is busy and cannot obtain the information, the user can try to obtain it again, or the user can re-enter new requirements to construct a new inspection and maintenance task.

[0040] If important inspection equipment information is successfully obtained, all inspection information will be displayed to the user, including inspection equipment, inspection time, inspection location, inspection priority, inspection type, execution method, etc. The user can change the above information by interactively entering text here. If there are any adjustments, the system will intelligently save the adjusted information here and still display the inspection information here. If there is no need to update, the inspection will proceed.

[0041] Input the updated inspection requirements into the code generation agent to obtain the inspection and maintenance script. Then, input the script into the maintenance expert agent and return the inspection results and suggestions. The inspection expert agent then retrieves and displays the results from the memory module to the user as the final inspection result.

[0042] (2) Process for handling requests unrelated to inspection The information entered is unrelated to inspection and maintenance work. It may be due to incorrect input or the possibility of casual conversation. Therefore, it is sufficient to simply reply "Please pay attention to the conversation" and ask the user to re-enter their requirements so that the process can start from the beginning again.

[0043] (3) Demand processing flow including weather conditions For the input requirement text, the inspection expert agent first uses a large language model to obtain the current time of the input text, which is still in the YYYY-MM-DD format. At the same time, it also obtains the location information and directly matches the initial input requirement text with the time information obtained in the previous step into the LTM long-term memory module of the Memory module. Then, it uses a large language model to obtain inspection-related suggestions. Since weather information is included here, the suggestions here are the areas and equipment to focus on under the current weather conditions.

[0044] Based on the obtained location and time information, the tool is invoked to use the Agent to obtain local equipment and weather information. Then, the obtained weather and equipment information is matched with the equipment and weather information obtained by the tool using the Agent and stored in STM to obtain the substation location and equipment information that matches the current weather conditions. If the tool uses the Agent to access the field equipment and times out, the processing procedure is the same as the "processing flow for requirements without weather information".

[0045] The inspection recommendations, information on important inspection equipment, and subsequent processing are consistent with the "processing flow for needs without weather information," ultimately yielding inspection results and recommendations.

[0046] (4) Processing procedures for comprehensive inspection Since it is necessary to detect all devices in the scene, after obtaining the requirement text, it is only necessary to calculate the current time. There is no need to match regional device suggestions, etc. Therefore, after inputting the requirement text, the patrol agent directly calculates the current time and calls the tool to use the agent to obtain device information of all locations.

[0047] The inspection content is summarized using a large language model, including inspection equipment, inspection time, inspection location, inspection priority, inspection type, and execution method. Users can interactively input text here to modify the above information. If adjustments are made, the system will intelligently save the adjusted information and still display the inspection information here. If there is no update requirement, the inspection will proceed. The subsequent overall processing flow is consistent with the "processing flow for requirements without weather information".

[0048] In this embodiment, the tool's Agent primarily functions to call on-site equipment for data collection. Based on the needs of the inspection task, it can determine weather conditions before inspection and maintenance, identify and schedule suitable physical equipment, and return equipment status and weather information to the Inspection Agent. Simultaneously, it collects image or sensor data in real time during the task and returns the results to the Maintenance Expert Agent. The tool's Agent includes a hardware control interface layer, a protocol conversion component, and a scene adaptation strategy layer. It can adapt protocols with cameras, infrared meters, drones, and environmental monitors deployed at different sites and complete data collection tasks. It supports cross-site equipment scheduling and provides a unified hardware call interface for multiple sites. Its main functions and processes in step S2 are as follows: 1. On-site equipment query and retrieval Based on the needs of the inspection expert agent, all equipment in the target substation is attempted to be called. For equipment that can be called, it is proven to be working normally, and the equipment availability information is returned to the inspection expert agent. This allows the inspection expert agent to determine whether the current substation is inspectable and whether manual repair is required, and the results are saved to STM.

[0049] 2. On-site information return The system calls upon on-site sensors to determine the weather conditions at the target power station, returns the results to the inspection expert agent for further assessment and processing, and saves the results to the STM.

[0050] In this embodiment, the main function of the Memory module is to realize information sharing and storage among intelligent agents, ensuring that information at each stage can be called. At the same time, it can also serve as a knowledge base to save inspection and maintenance-related knowledge, which facilitates subsequent summary and analysis of the results. It mainly includes an STM short-term memory module and an LTM long-term memory module, which respectively save the detection information of the current detection work, as well as the relevant knowledge and accumulated experience of the substation detection work as a whole. In step S2, the functions and processes of each module are as follows.

[0051] 1. STM Save the information on the field equipment, local weather conditions, and specific information about the field substation returned by the tool using the Agent.

[0052] 2. LTM The system predefines a large number of fault scenarios that may occur under different weather conditions, as well as suggestions and knowledge bases related to inspection and maintenance, to facilitate providing users with initial inspection suggestions.

[0053] The inspection information issued by the inspection expert agent, as well as the field equipment information and detection model version collected and fed back in STM, are saved for subsequent inspection and debugging.

[0054] Therefore, based on the above, in step S2 of this embodiment, when the inspection expert Agent calls the tool to obtain information about the power plant that matches the structured information content and generates inspection and maintenance suggestions, the specific steps include: S21) If the structured information includes time information, location information and weather information, the inspection expert Agent will use the time information and location information as the target time and target location, and call the tool to use the Agent to obtain the equipment information and weather information of all power stations at the target time and target location; The tool uses the Agent to attempt to call the equipment of all power stations at the target time and target location to obtain the equipment information of available equipment. The equipment information of each power station is returned to the inspection expert Agent and saved to the short-term memory module (STM). The tool also uses the Agent to call the field sensors of all power stations at the target time and target location to obtain weather information. The equipment availability information of each power station is returned to the inspection expert Agent and saved to the short-term memory module (STM). The inspection expert agent matches the weather information in the structured information with the weather information of all power stations at the target time and location stored in the short-term memory module (STM) to obtain the location and equipment information of the substation that matches the current weather conditions. The inspection expert agent combines the time and location information of the substation that matches the current weather conditions with the inspection and maintenance requirements input by the user. It matches the corresponding information in the long-term memory module (LTM). Based on the matching results, it uses a language model to generate inspection suggestions according to system prompts. The inspection suggestions include key areas and equipment. Then, based on the inspection suggestions, the time, location, and weather information in the structured information, and the equipment information stored in the short-term memory module (STM), it attempts to obtain the key equipment information to be focused on under the current weather conditions. S22) If the structured information contains time information and location information, the inspection expert Agent will use the time information and location information as the target time and target location, and call the tool to use the Agent to obtain the equipment information and weather information of all power stations at the target time and target location; The tool uses the Agent to attempt to call the equipment of all power stations at the target time and target location to obtain the equipment information of available equipment. The equipment information of each power station is returned to the inspection expert Agent and saved to the short-term memory module (STM). The tool also uses the Agent to call the field sensors of all power stations at the target time and target location to obtain weather information. The equipment availability information of each power station is returned to the inspection expert Agent and saved to the short-term memory module (STM). The inspection expert agent combines weather information with the user's inspection and maintenance needs, matches the corresponding information in the long-term memory module (LTM), and provides inspection suggestions using a language model based on the matching results and system prompts. The inspection suggestions include key areas and equipment to focus on. Then, by combining the inspection suggestions with the previously obtained location and time information, as well as the equipment information of each power station stored in the short-term memory module (STM), the agent attempts to obtain the key equipment information of the power station at the current location.

[0055] In steps S21 and S22, if the system is busy and cannot obtain information on key equipment, the user can try to obtain it again, or the user can re-enter new requirements to create a new inspection and maintenance task. If the information on important inspection equipment is successfully obtained, all inspection information will be displayed to the user, including the inspection equipment, inspection time, inspection location, inspection priority, inspection type, execution method, etc. The user can interactively input text here to modify the above information. If there are adjustments, the system will intelligently save the adjusted information here and still display the inspection information here. If there are no updated requirements, the inspection will proceed.

[0056] S23) If the structured information includes information related to the comprehensive inspection, the inspection expert Agent calculates the current time and calls the tool to use the Agent to obtain the equipment information of all power plants at the current time; The tool uses the Agent to attempt to call the equipment of all power stations at the current time, obtain the equipment information of available equipment, return the equipment information of each power station to the inspection expert Agent and save it in the short-term memory module STM; The inspection expert agent combines the user's inspection and maintenance needs with the corresponding information from the Long Memory Module (LTM). Based on the matching results and system prompts, it uses a language model to summarize the inspection content, including the inspected equipment, inspection time, inspection location, inspection priority, inspection type, and execution method. Users can interactively input text here to modify this information. If adjustments are made, the system will intelligently save the updated information and display it here; otherwise, the inspection will proceed.

[0057] S24) If the structured information contains information unrelated to inspection and maintenance work, the inspection expert agent will prompt the user to re-enter the requirements through the interactive page.

[0058] In this embodiment, when the code generation agent automatically generates the detection execution script, it automatically generates an execution script adapted to the detection engine based on information such as equipment, model type, and site environment in the inspection and maintenance plan. The script supports multi-type model calls, threshold configuration, and device scheduling command generation. Specifically, it obtains the corresponding model configuration and version from the long-term memory module (LTM) according to the task plan (location, equipment type, priority), dynamically generates an execution script adapted to the current task, and accumulates the script templates used in long-term operation, and optimizes the script generation logic. The main execution flow and functions in step S3 are as follows.

[0059] 1. Execute the script to generate The inspection requirements obtained by the inspection expert agent will be input into the code generation agent. The purpose is to generate an executable script to guide the subsequent maintenance expert agent to carry out the inspection work. The inspection requirements will include the inspection equipment, inspection time, inspection location, inspection priority, inspection type, etc.

[0060] The code generation agent first receives the inspection request from the inspection expert agent and converts the input content into a suitable format for subsequent use in the code generation process.

[0061] Based on the inspection location and inspection time, the tool uses the Agent to query the local hardware devices and detection model versions available at that moment.

[0062] Based on the script template for the current model and information such as the inspection type, an execution script is generated, and the script parameters are dynamically adjusted according to information such as the inspection equipment and inspection priority.

[0063] 2. Optimize script logic The code generation process is based on the corresponding model and predefined script templates. Initially, these predefined scripts were added manually. However, as the number of connected power stations and model versions increases, the templates need to be updated. During the initial manual updates, it's common to add corresponding script generation templates for different model versions. However, if there are too many models, this can lead to increased code volume, redundancy, and difficulty in finding relevant information. The code generation agent applies a large language model to find common interfaces and parameters among the existing script templates and optimizes and unifies them. For example, if only a few parameters differ between the script templates for two models, the code generation agent will set the changed parameters as optional, allowing a unified template to be used for multiple models when generating scripts subsequently.

[0064] Correspondingly, the main functions and processes of the tool using the Agent in step S3 are as follows: Based on the requirements for generating the Agent, query the detection model and corresponding version used by the substation at the target location, and return the model information so that the script can be generated.

[0065] Therefore, based on the above, in step S3 of this embodiment, after the code generation agent obtains the corresponding model configuration and version from the long-term memory module (LTM) of the memory module to generate the execution script, it also saves the corresponding script template. After dynamically adjusting the script parameters according to the information of the inspection and maintenance tasks, the code generation agent also applies a large language model to find common interfaces and common parameters for the generated execution script templates and optimizes and unifies them to obtain a unified template representation.

[0066] In this embodiment, the maintenance expert agent is equipped with a defect diagnosis module based on inspection and maintenance data. It locates faults based on image recognition results, task configuration priorities, and short-term memory (STM) content, and calls the repair knowledge base to generate multiple solution recommendations. Its main functions are to receive task plans issued by the superior agent, execute equipment detection based on image data, analyze the recognition results, determine the equipment status and provide maintenance suggestions, and write the detection results into the STM of the memory module in real time for strategy optimization in the current task stage. Finally, after the inspection and maintenance are completed, the key features in the STM can be summarized and this information can be updated in the LTM. The main functions and processes in step S4 are as follows.

[0067] 1. Equipment defect detection The code generates the execution script information passed in by the Agent, initializes and runs the on-site detection model, and performs detection based on the passed parameters such as the device category to be identified and the detection threshold.

[0068] The tool uses an agent to input video information or video streams. The model detects the video or video stream and returns the detected results, which are then saved to the STM in the Memory module.

[0069] 2. Maintenance suggestion generation and information summary During the inspection process, the current inspection results are saved to the STM, and the main problems are summarized at regular intervals. This summary is still based on the maintenance knowledge saved in the LTM, and content for adjusting the training strategy is generated according to the prompts. The inspection strategy is then dynamically adjusted accordingly. The system prompt here is: "You are a maintenance technician with many years of work experience. Based on the defects found so far, combined with the maintenance knowledge you have found, please summarize the main problems that have occurred and provide the optimal maintenance strategy, in a format similar to maintenance equipment - inspection priority - inspection type." Subsequently, the maintenance expert agent will read the modified strategy content and change the inspection strategy accordingly.

[0070] After the inspection is completed, the model will save all the detected results to the STM. The maintenance expert agent will query all relevant information in the STM and apply the large language model, combined with the maintenance knowledge information stored in the LTM, to give comprehensive inspection suggestions. The system prompt for the large language model is: "You are an expert specializing in substation maintenance. The user asked about the above defects and gave corresponding maintenance suggestions and future precautions. Please answer according to the knowledge queried below."

[0071] STM is a short-term memory module that only stores information for the current task. In order to update the overall knowledge, after the inspection work is completed, the maintenance expert agent will call the large language model to summarize the inspection results in STM and pass them into LTM for reference in subsequent inspection work.

[0072] Correspondingly, the main functions and processes of the tool using the Agent in step S4 are as follows: It can call the devices at the target site, such as on-site surveillance cameras, and return video data for the whole day or a whole time period. For ongoing maintenance work, it can return the current video stream information, which can facilitate subsequent inspection work by the maintenance expert agent.

[0073] Correspondingly, the main functions and processes of each module in the Memory module in step S4 are as follows: 1. STM Save the inspection results returned by the maintenance expert agent each time, using the template format: "{Time, Equipment, Defect Name, Inspection Frame Coordinates, Total Number of Inspections for this Defect, Inspection Priority}". This saves a summary of the current detection results compiled by the large language model. To prevent confusion with the detection structure, it is saved as a dictionary. A simple sample template is: "{Result Summary: Under the current weather conditions, the expander is prone to overshooting; please prioritize identifying this type of problem}". 2. LTM The system compiles information on current environmental conditions, equipment defects, and precautions from the STM and imports it into the LTM, updating the knowledge base to keep the data and experience up-to-date.

[0074] Therefore, based on the above, in step S4 of this embodiment, when the maintenance expert agent provides maintenance suggestions and adjusts the detection strategy based on the detection results in the short-term memory module (STM) and the maintenance knowledge information stored in the long-term memory module (LTM), the specific steps include: The maintenance expert agent uses a language model to generate content for adjusting the training strategy based on the maintenance knowledge information stored in the long-term memory module (LTM) and the detection results in the short-term memory module (STM) according to a specified period. It also reads the latest strategy content and modifies the detection strategy accordingly. The maintenance expert agent queries all relevant information in the short-term memory module (STM) and applies a large language model, combining it with maintenance knowledge information stored in the long-term memory module (LTM) to provide comprehensive maintenance recommendations.

[0075] Furthermore, in step S4 of this embodiment, after saving the detection results to the Short-Term Memory (STM) module of the memory module, the method further includes: after the detection work is completed, the maintenance expert agent calls the large language model to summarize the detection results in the STM module and transmits them to the Long-Term Memory (LTM) module, thereby updating the maintenance knowledge information stored in the LTM module for reference in subsequent detection work.

[0076] Furthermore, in step S4 of this embodiment, after the maintenance expert agent provides maintenance suggestions and adjusts the detection strategy based on the detection results in the short-term memory module (STM) and the maintenance knowledge information stored in the long-term memory module (LTM), the maintenance expert agent also provides feedback on the maintenance suggestions to the inspection expert agent, and the inspection expert agent displays the maintenance suggestions through an interactive page.

[0077] The workflow of the system in this embodiment will be further explained below with a specific example.

[0078] First, the inspection expert agent receives the user's input request text. Then, it uses a large language model to extract text information and determine the type of information. For example, if the input is "Hello, how tall are you?", the inspection expert agent will prompt "Pay attention to the input content" and allow the user to re-enter the request.

[0079] If the user re-enters their request text, such as "I want to inspect the operation of the Shenzhen Longhua Substation on July 18, 2025. Which equipment should I focus on?", the AI ​​will detect that the current input does not include weather information and is a reasonable inspection request. It will then extract the time and location information: July 18, 2025 and Shenzhen Longhua Substation.

[0080] The time and location information is transmitted from the inspection expert agent to the tool's agent to obtain the status of various monitoring and sensor data at the Shenzhen Longhua substation up to July 18, 2025. This allows the tool to determine which devices can be used to obtain information in the future and to query the detection models and version information deployed at the substation.

[0081] The tool uses an agent to call sensors to determine the weather conditions on site, and at the same time calls the monitoring equipment to return video stream information. It then transmits the weather and equipment status information back to the inspection expert agent and saves the information to the STM short-term memory module of the memory module as information for this inspection task.

[0082] The inspection expert Agent will call the inspection knowledge stored in the long-term memory module (LTM) in the Memory module, match the existing equipment and weather information with the inspection knowledge, and use a large language model to sort out the work suggestions for inspection and maintenance process corresponding to the current weather conditions and site equipment conditions.

[0083] Based on the work suggestions and the initial inspection and maintenance requirements, an inspection task is generated and the task information is displayed, including the inspection equipment, inspection time, inspection location, inspection priority, inspection type, and execution method.

[0084] Users can check the above information and choose whether to change it according to the latest task requirements. If changes are required, users can enter the information category to be changed. For example, if the inspection time is to be changed, users can enter "Inspection: 2025-07-05". Then, the inspection and maintenance suggestions will be regenerated based on the new requirements, and the inspection task will be generated again. The system will then determine whether changes are needed until it is confirmed that no changes have been made.

[0085] If the inspection task is correct, the task information will be sent to the code generation agent. The agent will read the relevant information. Considering that the information may have changed at this time, the code generation agent will call the tool again to use the agent to obtain the current model configuration and field equipment information, and formulate a strategy based on the current equipment and model status to generate an executable script.

[0086] The executable script is input into the maintenance expert agent. The agent builds a detection model based on the script content, receives video stream data from the tool using the agent for detection, continuously transmits the detection results to the STM, and periodically uses the large language model to summarize the output results and adjust the maintenance strategy.

[0087] After the inspection is completed, all inspection results will be returned. The maintenance expert agent will combine the inspection results to generate maintenance suggestions and feed them back to the inspection expert agent. The inspection expert agent will then display the results to the user through an interactive page.

[0088] In summary, this invention proposes a multi-agent-based cross-site intelligent inspection and maintenance method and system. It achieves closed-loop automated processing of the entire process—from natural language task parsing, equipment scheduling, data acquisition, fault detection, and intelligent maintenance suggestions—through inspection expert agents, maintenance expert agents, tool-using agents, and code generation agents. Furthermore, it introduces short-term memory (STM) and long-term memory (LTM) to support experience accumulation and knowledge reuse. The inspection expert agent is responsible for task planning and multi-agent coordination; the maintenance expert agent performs status judgment and strategy optimization based on image or sensor data; the tool-using agent implements multi-source equipment protocol adaptation and data acquisition; and the code generation agent automatically generates detection scripts based on environmental conditions. This invention can be applied to intelligent inspection and maintenance scenarios for facilities such as substations with multiple sites and complex climatic conditions, achieving efficient, intelligent, and sustainable automated task management and execution.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A cross-site intelligent inspection and maintenance method based on multi-agent systems, characterized in that, The method is applied to a cross-site intelligent inspection and maintenance system that includes an inspection expert agent, a tool-using agent, a code-generating agent, a maintenance expert agent, and a memory module. The method includes the following steps: The inspection expert intelligent agent receives the inspection and maintenance requirements input by the user, performs semantic parsing through a large language model, extracts structured information, then calls the tool to use the intelligent agent to obtain the power station information that matches the structured information content and generates inspection and maintenance suggestions. Finally, the corresponding power station equipment status and inspection and maintenance suggestions are displayed through the interactive page, and the corresponding inspection and maintenance tasks are issued after the user confirms. The code generation agent receives inspection and maintenance tasks, calls tools to use the agent to obtain the model version information and hardware equipment information of the corresponding power station, and then, based on the model version information and hardware equipment information, retrieves the corresponding model configuration and version from the long-term memory module of the memory module to generate an execution script, and dynamically adjusts the script parameters according to the information of the inspection and maintenance tasks. The maintenance expert agent obtains the execution script, initializes and runs the detection model of the power plant site corresponding to the inspection and maintenance task according to the execution script information, calls the tool to use the agent to obtain the monitoring information of the corresponding power plant site, uses the detection model to detect the monitoring information, and saves the detection results to the short-term memory module of the memory module. The maintenance expert agent provides maintenance suggestions and adjusts the detection strategy based on the detection results in the short-term memory module and the maintenance knowledge information stored in the long-term memory module.

2. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, The structured information includes time information, location information, and weather information. When the inspection expert agent calls the tool to obtain information about the power station that matches the structured information and generates inspection and maintenance suggestions, it specifically includes: The inspection expert intelligent agent uses the time and location information as the target time and target location, and calls the tool to use the intelligent agent to obtain the equipment information and weather information of all power stations at the target time and target location; The tool uses an intelligent agent to attempt to call the equipment of all power stations at the target time and target location to obtain the equipment information of available equipment. It returns the equipment information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The tool also uses an intelligent agent to call the field sensors of all power stations at the target time and target location to obtain weather information. It returns the equipment availability information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The inspection expert intelligent agent matches the weather information in the structured information with the weather information of all power stations at the target time and target location stored in the short-term memory module to obtain the location and equipment information of the substation that matches the current weather conditions; The inspection expert agent combines the substation's time and location information, which are consistent with the current weather conditions, with the inspection and maintenance requirements input by the user. It matches the corresponding information in the long-term memory module. Based on the matching results and system prompts, it uses a language model to generate inspection suggestions. The inspection suggestions include key areas and equipment to focus on. Then, by combining the inspection suggestions with the time, location, and weather information in the structured information, and the equipment information stored in the short-term memory module, it attempts to obtain the key equipment information to focus on under the current weather conditions.

3. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, The structured information includes time and location information. When the inspection expert agent calls the tool to obtain information about the power station that matches the structured information and generates inspection and maintenance suggestions, it specifically includes: The inspection expert intelligent agent uses the time and location information as the target time and target location, and calls the tool to use the intelligent agent to obtain the equipment information and weather information of all power stations at the target time and target location; The tool uses an intelligent agent to attempt to call the equipment of all power stations at the target time and target location to obtain the equipment information of available equipment. It returns the equipment information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The tool also uses an intelligent agent to call the field sensors of all power stations at the target time and target location to obtain weather information. It returns the equipment availability information of each power station to the inspection expert intelligent agent and saves it in the short-term memory module. The inspection expert intelligent agent combines weather information with the user's inspection and maintenance needs, matches the corresponding information in the long-term memory module, and provides inspection suggestions using a language model based on the matching results and system prompts. The inspection suggestions include key areas and equipment to focus on. Then, by combining the inspection suggestions with the previously obtained location and time information, as well as the equipment information of each power station stored in the short-term memory module, it attempts to obtain the key equipment information of the power station at the current location.

4. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, The structured information includes information related to the comprehensive inspection. When the inspection expert agent invokes tools to obtain power plant information matching the structured information content and generates inspection and maintenance suggestions, it specifically includes: The inspection expert agent calculates the current time and calls upon tools to obtain equipment information for all power plants at the current time. The tool uses an intelligent agent to attempt to call the equipment of all power stations at the current time, obtain the equipment information of available equipment, return the equipment information of each power station to the inspection expert intelligent agent and save it to the short-term memory module; The inspection expert intelligent agent combines the user's inspection and maintenance needs with the corresponding information in the long-term memory module. Based on the matching results, it uses a language model to summarize the inspection content, including the inspection equipment, inspection time, inspection location, inspection priority, inspection type, and execution method, along with system prompts.

5. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, The structured information includes information unrelated to inspection and maintenance work. When the inspection expert agent calls the tool to obtain power plant information that matches the structured information content and generates inspection and maintenance suggestions, the specific steps include: If the structured information contains information unrelated to inspection and maintenance work, the inspection expert intelligent agent will prompt the user to re-enter the requirements through an interactive page.

6. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, After the code generation agent obtains the corresponding model configuration and version from the long-term memory module of the memory module to generate the execution script, it also saves the corresponding script template. After dynamically adjusting the script parameters according to the information of inspection and maintenance tasks, it also includes: the code generation agent applying a large language model to find common interfaces and common parameters for the generated execution script templates, and optimizing and unifying them to obtain a unified template representation.

7. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, When the maintenance expert agent provides maintenance suggestions and adjusts the testing strategy based on the test results in the short-term memory module and the maintenance knowledge information stored in the long-term memory module, it specifically includes: The maintenance expert agent uses a language model to generate content for adjusting the training strategy based on the maintenance knowledge information stored in the long-term memory module and the test results in the short-term memory module, according to a specified period. It also reads the latest strategy content and modifies the test strategy accordingly. The maintenance expert agent queries all relevant information in the short-term memory module and applies a large language model, combined with maintenance knowledge information stored in the long-term memory module, to provide comprehensive maintenance suggestions.

8. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, After saving the test results to the short-term memory module of the memory module, the process also includes: after the test is completed, the maintenance expert agent calls the large language model to summarize the test results in the short-term memory module and transfers them to the long-term memory module, thereby updating the maintenance knowledge information stored in the long-term memory module for reference in subsequent test work.

9. The cross-site intelligent inspection and maintenance method based on multi-agent systems according to claim 1, characterized in that, After the maintenance expert agent provides maintenance suggestions and adjusts the detection strategy based on the detection results in the short-term memory module and the maintenance knowledge information stored in the long-term memory module, the maintenance expert agent also provides feedback on the maintenance suggestions to the inspection expert agent, which then displays the maintenance suggestions through an interactive page.

10. A multi-agent-based cross-site intelligent inspection and maintenance system, characterized in that, The system includes an inspection expert agent, a tool-using agent, a code-generating agent, a maintenance expert agent, and a memory module. The multi-agent-based cross-site intelligent inspection and maintenance system is programmed or configured to execute the steps of the multi-agent-based cross-site intelligent inspection and maintenance method as described in any one of claims 1 to 9.

Images