Dual-mode linkage control method and system for intelligent inspection of power equipment

Abstract

The embodiment of the invention relates to the technical field of intelligent monitoring, and discloses a dual-mode linkage control method, system and device for intelligent inspection of power equipment and a storage medium, and the method comprises the steps: issuing an inspection task through a decision-making agent; the conventional inspection agent performs fine adjustment on the inspection task, inspects each piece of power equipment in the power system, performs real-time initial analysis according to inspection data, and sends a real-time initial analysis result to the decision agent in real time; the decision-making agent performs analysis according to the real-time initial analysis result of the conventional inspection agent to obtain a target analysis result and determines a detailed inspection mode task; the detailed investigation intelligent agent carries out detailed investigation inspection at the target power equipment, carries out initial detailed investigation analysis on detailed investigation data, and sends the data to the decision intelligent agent; and the decision-making agent performs analysis according to the detailed investigation data and the initial detailed investigation analysis to obtain a target detailed investigation analysis result, and performs early warning according to the target detailed investigation analysis result. According to the embodiment of the invention, dual-mode intelligent linkage of conventional inspection and detailed inspection is realized, and comprehensive analysis and adaptive scheduling can be realized.

Description

TECHNICAL FIELD

[0001] The embodiment of the present application relates to the technical field of intelligent monitoring, in particular to a dual-mode linkage control method and system for intelligent inspection of power equipment. BACKGROUND

[0002] At present, with the rapid development of smart grid, the scale of power equipment is increasingly large, and higher requirements are put forward for the inspection and maintenance of its operating state. The traditional inspection mode mainly relies on manual or single inspection robot, and there are problems of low efficiency, high risk, and inability to respond in real time.

[0003] In the prior art, some schemes for automatic inspection by inspection robots have appeared. However, these schemes mostly have the following defects: 1. Single function: The conventional inspection robot usually only performs fixed cruising tasks, the collected data is simple, the analysis ability is weak, and the suspected defects cannot be deeply probed.

[0004] 2. Poor linkage: The conventional inspection and detailed inspection are disconnected. Artificial intervention is usually needed to determine whether detailed inspection is needed, and the task is dispatched again, which is slow in response and easy to miss the processing opportunity.

[0005] 3. Insufficient intelligence level: The decision-making process mainly depends on the preset fixed threshold, lacks comprehensive cognition and reasoning ability for multi-source information, device topology and historical data, has high false alarm rate, and cannot predict the development trend of defects.

[0006] 4. Weak system robustness: The whole system depends on the stable operation of each intelligent agent, and lacks an emergency task allocation mechanism when a single intelligent agent fails. SUMMARY

[0007] In view of the above problems, the embodiment of the present application provides a dual-mode linkage control method and system for intelligent inspection of power equipment, a device and a storage medium, which are used to solve the problems of single function, inability to link the dual-mode intelligent inspection of conventional inspection and detailed inspection, comprehensive analysis and adaptive scheduling in the prior art.

[0008] According to an aspect of the embodiment of the present application, a dual-mode linkage control method for intelligent inspection of power equipment is provided, which comprises: The method is based on a plurality of power inspection intelligent agents, and the power inspection intelligent agents comprise a decision-making intelligent agent, a conventional inspection intelligent agent and a detailed inspection intelligent agent; the method comprises: issuing an inspection task by the decision-making intelligent agent; the inspection task comprises a conventional cruising mode task; the conventional cruising mode task comprises a conventional route, an inspection frequency and an inspection content of the conventional cruising mode task; The conventional inspection agent fine-tunes the inspection task according to the conventional cruise mode task, a real-time operation mode of the power grid, and meteorological environment information, inspects each power device in the power system according to the fine-tuned inspection task, obtains inspection data, performs real-time initial analysis according to the inspection data, and sends a real-time initial analysis result to the decision agent in real time; the real-time initial analysis result includes an abnormal power device identifier, an abnormal type, and a confidence level; The decision agent performs comprehensive analysis to obtain a target analysis result according to the real-time initial analysis result sent by each conventional inspection agent, the historical operation and maintenance record of the abnormal power device, the current power grid operation condition, and the power device knowledge base information, and determines a detailed inspection mode task according to the target analysis result; The detailed inspection agent performs detailed inspection according to the detailed inspection mode task to the target power device, and sends an initial detailed inspection analysis to the decision agent after performing initial detailed inspection analysis on the detailed inspection data; The decision agent obtains a target detailed inspection analysis result after performing analysis on the detailed inspection data and the initial detailed inspection analysis, and performs early warning according to the target detailed inspection analysis result.

[0009] In an optional manner, before the decision agent issues the inspection task, the method further includes: The decision agent obtains power device information and power device distribution information of each power device in the power system from a device information base of the power system; the power device information includes power device types, physical parameters, and electrical parameters; and the power device distribution information includes geographical positions of each power device and electrical connection relationships between the power devices; The decision agent inputs the power device information and the power device distribution information of each power device in the power system into the fine-tuned large model to obtain the conventional cruise mode task.

[0010] In an optional manner, before the decision agent performs comprehensive analysis to obtain a target analysis result according to the real-time initial analysis result sent by each conventional inspection agent, the historical operation and maintenance record of the abnormal power device, the current power grid operation condition, and the power device knowledge base information, and determines a detailed inspection mode task according to the target analysis result, the method includes: Generating detailed inspection training data according to the power device information, the power device distribution information, the historical operation and maintenance data, the counter-operation and maintenance data, and the expert experience operation and maintenance data of the power grid system, and corresponding detailed inspection labels; Converting the detailed inspection training data into context prompt data, and fine-tuning the large model by using LoRA to obtain a fine-tuned large model; Input the detailed investigation training data into a lightweight neural network for training to obtain a detailed investigation prediction model; According to the power equipment information, power equipment distribution information, historical operation and maintenance data, counter operation and maintenance data and expert experience operation and maintenance data of the power grid system, a dynamic anomaly knowledge graph is generated; the dynamic anomaly knowledge graph includes the topological relationship between power equipment, historical anomaly records, and defect evolution law.

[0011] In an optional manner, the decision agent performs comprehensive analysis according to the real-time initial analysis result sent by each regular inspection agent, the historical operation and maintenance record corresponding to the abnormal power equipment, the current power grid operation condition and the power equipment knowledge base information to obtain a target analysis result, and determines a detailed investigation and inspection mode task according to the target analysis result, including: Input the real-time initial analysis result, the historical operation and maintenance record corresponding to the abnormal power equipment, and the current power grid operation condition into a detailed investigation prediction model to obtain a first detailed investigation task; Input the power equipment information, power equipment distribution information, historical operation and maintenance data, counter operation and maintenance data and expert experience operation and maintenance data of the power grid system into the fine-tuned large model to obtain a second detailed investigation task; Input the real-time initial analysis result and the current power grid operation condition into the dynamic anomaly knowledge graph to obtain a third detailed investigation task; Determine the detailed investigation and inspection mode task according to the first detailed investigation task, the second detailed investigation task and the third detailed investigation task.

[0012] In an optional manner, the regular inspection agent fine-tunes the inspection task according to the regular cruise mode task, the real-time operation mode of the power grid and the meteorological environment information, and inspects each power equipment in the power system according to the fine-tuned inspection task to obtain inspection data, performs real-time initial analysis according to the inspection data, and sends the real-time initial analysis result to the decision agent in real time, including: When the regular inspection agent detects that the load rate of the target line or the target main transformer in the power system exceeds the preset load threshold, the inspection priority of the target line or the target main transformer is increased; When the regular inspection agent determines that the target power equipment is in a power-off maintenance state through the switch state, the target power equipment is automatically skipped in the current regular cruise mode task, and the state is fed back to the decision agent; When the regular inspection agent detects that the meteorological environment corresponding to the current inspection area is lower than the first environment threshold and the regional line load exceeds the preset load area, the inspection route is adjusted; When the regular inspection agent detects that the meteorological environment corresponding to the current inspection area is lower than a second environment threshold, stop the inspection and feed back the environment information to the decision agent; the first environment threshold is higher than the second environment threshold.

[0013] In an optional mode, the decision agent performs early warning according to the target detailed analysis result, and specifically includes: The decision agent compares the target detailed analysis result with a plurality of preset early warning thresholds of different levels; According to the comparison result, different levels of early warning responses are triggered; the different levels of early warning responses at least include a first level early warning, a second level early warning and a third level early warning; When the first level early warning is triggered, a maintenance work order is generated and is included in the planned maintenance process; When the second level early warning is triggered, alarm information is sent to the regional operation and maintenance personnel, and feedback is required within a specified time; When the third level early warning is triggered, alarm information is sent to the regional operation and maintenance personnel, and an emergency alarm is sent to the dispatch center and the control system, and an emergency operation strategy is recommended, and the corresponding power equipment is automatically linked and isolated.

[0014] In an optional mode, the inspection data at least includes visible light images and infrared temperature data of the power equipment; the regular inspection agent inspects the power equipment, obtains the inspection data, and performs real-time initial analysis according to the inspection data, and specifically includes: The regular inspection agent synchronously collects visible light images and infrared temperature data of the power equipment through the visible light camera and the infrared thermal imager carried during the cruise; The regular inspection agent performs real-time initial analysis through the lightweight neural network model deployed locally, to identify meter reading, mechanical deformation and foreign object attachment from the visible light images, and to identify temperature abnormal points from the infrared temperature data; wherein the real-time initial analysis result is a structured data object, at least including equipment ID, abnormal type, abnormal position coordinates, abnormal confidence and time stamp.

[0015] According to another aspect of the embodiment of the application, a dual-mode linkage control system for intelligent inspection of power equipment is provided, the system includes a plurality of power inspection agents, the power inspection agents include a decision agent, a regular inspection agent and a detailed inspection agent; wherein: The decision agent is configured to issue an inspection task; the inspection task includes a regular cruise mode task; the regular cruise mode task includes a regular route, an inspection frequency and an inspection content of the regular cruise mode task; The conventional inspection agent is configured to fine-tune an inspection task according to the conventional cruise mode task, a real-time operation mode of a power grid, and meteorological environment information, perform inspection on each power equipment in the power system according to the fine-tuned inspection task, obtain inspection data, perform real-time initial analysis according to the inspection data, and send a real-time initial analysis result to the decision agent in real time; the real-time initial analysis result includes an abnormal power equipment identifier, an abnormal type, and a confidence level. The decision agent is further configured to perform comprehensive analysis to obtain a target analysis result according to the real-time initial analysis result sent by each conventional inspection agent, the real-time initial analysis result, historical operation and maintenance records of the abnormal power equipment, a current power grid operation condition, and power equipment knowledge base information, and determine a detailed inspection mode task according to the target analysis result. The detailed inspection agent is configured to perform detailed inspection on the target power equipment according to the detailed inspection mode task, and send a detailed inspection result to the decision agent after performing initial detailed inspection analysis on the detailed inspection data. The decision agent is further configured to obtain a target detailed inspection analysis result after performing analysis on the detailed inspection data and the initial detailed inspection analysis, and perform early warning according to the target detailed inspection analysis result.

[0016] In an optional manner, the decision agent is further configured to: generate detailed inspection training data according to each power equipment information, power equipment distribution information, historical operation and maintenance data, adversarial operation and maintenance data, and expert experience operation and maintenance data of the power grid system, and corresponding detailed inspection labels; convert the detailed inspection training data into context prompt data, and fine-tune a large model using LoRA to obtain a fine-tuned large model; input the detailed inspection training data into a lightweight neural network to train a detailed inspection prediction model; generate a dynamic abnormal knowledge graph according to each power equipment information, power equipment distribution information, historical operation and maintenance data, adversarial operation and maintenance data, and expert experience operation and maintenance data of the power grid system; the dynamic abnormal knowledge graph includes a topological relationship between power equipment, historical abnormal records, and defect evolution rules.

[0017] In an optional manner, the decision agent is further configured to: input the real-time initial analysis result, the historical operation and maintenance records of the abnormal power equipment, and the current power grid operation condition into the detailed inspection prediction model to obtain a first detailed inspection task; input each power equipment information, power equipment distribution information, historical operation and maintenance data, adversarial operation and maintenance data, and expert experience operation and maintenance data of the power grid system into the fine-tuned large model to obtain a second detailed inspection task; input the real-time initial analysis result and the current power grid operation condition into the dynamic abnormal knowledge graph to obtain a third detailed inspection task; According to the first detailed inspection task, the second detailed inspection task and the third detailed inspection task, the detailed inspection and inspection mode task is determined.

[0018] The method provided by the embodiment of the application is based on a plurality of power inspection intelligent agents, including a decision intelligent agent, a regular inspection intelligent agent and a detailed inspection intelligent agent. The decision intelligent agent issues an inspection task; the inspection task includes a regular cruise mode task; the regular inspection intelligent agent fine-tunes the inspection task according to the regular cruise mode task, a real-time operation mode of a power grid and meteorological environment information, and inspects each power equipment in the power system according to the fine-tuned inspection task to obtain inspection data, performs real-time initial analysis according to the inspection data, and sends the real-time initial analysis result to the decision intelligent agent in real time; the decision intelligent agent performs comprehensive analysis according to the real-time initial analysis result sent by each regular inspection intelligent agent, the real-time initial analysis result, historical operation and maintenance records of the abnormal power equipment, a current power grid operation condition and power equipment knowledge base information to obtain a target analysis result, and determines a detailed inspection and inspection mode task according to the target analysis result; the detailed inspection intelligent agent performs detailed inspection to a target power equipment according to the detailed inspection and inspection mode task, and sends initial detailed inspection analysis to the decision intelligent agent after performing initial detailed inspection analysis on detailed inspection data; the decision intelligent agent obtains a target detailed inspection analysis result after performing analysis on the detailed inspection data and the initial detailed inspection analysis, and performs early warning according to the target detailed inspection analysis result. Through the dual-mode linkage control of the power equipment intelligent inspection, the dual-mode intelligent linkage of the regular inspection and the detailed inspection is realized, and comprehensive analysis and adaptive scheduling are achieved.

[0019] The above description is only a summary of the technical scheme of the embodiment of the application, in order to more clearly understand the technical means of the embodiment of the application, the embodiment of the application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the embodiment of the application more obvious and easy to understand, the specific implementation manner of the application is described below. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings are only used to show the embodiments and are not considered as limitations of the application. Moreover, the same reference signs are used to represent the same components throughout the drawings. In the drawings: Figure 1 A flowchart of a dual-mode linkage control method of power equipment intelligent inspection provided by the embodiment of the application is shown; Figure 2 A flowchart of a dual-mode linkage control method of power equipment intelligent inspection provided by another embodiment of the application is shown; Figure 3A structural schematic diagram of a dual-mode linkage control system for intelligent inspection of power equipment is shown. DETAILED DESCRIPTION

[0021] Exemplary embodiments of the present application will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein.

[0022] Figure 1 A flowchart of a dual-mode linkage control method for intelligent inspection of power equipment is shown, which is executed by a dual-mode linkage control system for intelligent inspection of power equipment, the system comprising a plurality of power inspection agents, wherein the power inspection agents include entities that can be fixed or mobile, and can perceive the environment, make decisions and take actions to achieve specific goals. For example, the power inspection agents can be mobile robots, unmanned aerial vehicles, etc. equipped with various sensors, fixed monitoring agents arranged in fixed areas, etc. For example, the fixed monitoring agents arranged in fixed areas can be fixed cameras, infrared ball machines, micro meteorological stations arranged on towers and structures. The ground mobile robots can be robots carrying high-definition pan-tilt cameras, infrared thermographs, ultrasonic / partial discharge detection sensors, and mechanical arms. The unmanned aerial vehicles can be multi-rotor unmanned aerial vehicles carrying wide-angle visible light cameras, medium-resolution infrared thermographs, and communication relay modules, etc.

[0023] As shown in Figure 2 From the perspective of inspection mode, the power inspection agents include one decision-making agent, a plurality of regular inspection agents and a plurality of detailed inspection agents, each group of regular inspection agents being responsible for inspecting power equipment in a specific area. As shown in Figure 1 The method comprises the following steps: Step 110: issuing an inspection task by the decision-making agent.

[0024] The decision-making agent, as a central agent, is the core decision-making center of the system, responsible for task allocation, macro decision-making and resource scheduling. In the embodiments of the present application, the inspection task includes a regular cruise mode task; the regular cruise mode task includes a regular route, an inspection frequency and an inspection content of the regular cruise mode task.

[0025] The decision-making agent obtains information of each power equipment in the power system and distribution information of the power equipment according to an equipment information database of the power system; the information of each power equipment includes types, physical parameters and electrical parameters of the power equipment; and the distribution information of the power equipment includes geographical positions of each power equipment and electrical connection relationships between the power equipment.

[0026] In the embodiment of the present application, the final routine cruise mode task is obtained through fine-tuning of the large model to determine the routine cruise mode task and then confirmed by an expert. The decision-making agent inputs the information of each power equipment in the power system and the distribution information of the power equipment into the fine-tuned large model to obtain the routine cruise mode task. Specifically, when generating the routine cruise mode task, the decision-making agent is no longer simply circulating according to a fixed route, but is based on a multi-dimensional decision-making model, in which the power equipment type and distribution are key inputs. The power equipment type is an intrinsic attribute, including different types of power equipment, whose fault mode, key monitoring parameters, inspection requirements and safety risks are completely different. For example, for the following power equipment, the following key monitoring parameters are included. Transformer: monitoring parameters include focusing on oil temperature, oil level, sleeve, cooling system and sound anomaly. Inspection needs to be close to collect infrared thermal image and audio data. Circuit breaker: focus on the on-off position, mechanism box sealing, SF6 gas pressure and infrared temperature measurement. Transmission line / insulator string: focus on foreign matter, icing, dancing, insulator damage and arc trace. Inspection needs to cover the entire length of the line. Surge arrester: focus on leakage current and action times. Power cable: focus on partial discharge, surface temperature and external damage risk on the path. Among them, the type of power equipment directly determines the type of sensor used, the inspection point and the inspection accuracy. The inspection accuracy includes inspection distance and resolution. For power equipment distribution information, it includes geographical distribution and electrical connection distribution. Among them, the geographical distribution refers to the physical location of the power equipment in two-dimensional or three-dimensional space. This is the basis for planning the optimal inspection path, aiming to minimize the empty distance and energy consumption of the inspection agent and maximize the coverage range of a single inspection. The electrical connection distribution (topology) refers to the connection relationship of the power equipment in the power primary system diagram, such as who is in series / parallel with whom, and who is the power supply side and who is the load side. This determines the fault impact range. The failure of a key node (such as the main transformer of a hub substation) has a much greater impact than that of an end device. Therefore, the inspection frequency and priority of key nodes should be higher. In the embodiment of the present application, the routine cruise mode task further includes a preset inspection list, a pre-defined inspection route, a target device sequence, and standard inspection items for each device.

[0027] Through the above fine-tuned large model to determine the routine cruise mode task, and then confirmed by an expert, the final routine cruise mode task is obtained. The routine cruise mode task is a fixed interval frequency cruise task, and the decision-making agent periodically determines the routine cruise mode task and issues a cruise instruction to each region of the power system.

[0028] Step 120: The regular inspection agent fine-tunes the regular cruise mode task according to the regular cruise mode task, the real-time operation mode of the power grid, and the meteorological environment information, and inspects each power equipment in the power system according to the fine-tuned inspection task, obtains inspection data, performs real-time initial analysis according to the inspection data, and sends the real-time initial analysis result to the decision agent in real time.

[0029] In the embodiment of the application, the regular inspection agent fine-tunes the inspection task according to the regular cruise mode task, the real-time operation mode of the power grid, and the meteorological environment information before inspection, in order to avoid temporary environmental and power equipment emergencies. The regular inspection agent in the embodiment of the application has a certain degree of edge intelligence, so that it can make local adjustments under the task framework formulated by the decision agent based on the dynamically changing environment and system state, thereby realizing safer, more efficient, and more accurate inspection.

[0030] In the embodiment of the application, the regular inspection agent fine-tunes the inspection task according to the regular cruise mode task, the real-time operation mode of the power grid, and the meteorological environment information before inspection, in order to avoid temporary environmental and power equipment emergencies. The regular inspection agent in the embodiment of the application has a certain degree of edge intelligence, so that it can make local adjustments under the task framework formulated by the decision agent based on the dynamically changing environment and system state, thereby realizing safer, more efficient, and more accurate inspection.

[0031] In the embodiment of the application, the regular inspection agent fine-tunes the inspection task according to the regular cruise mode task, the real-time operation mode of the power grid, and the meteorological environment information before inspection, in order to avoid temporary environmental and power equipment emergencies. The regular inspection agent in the embodiment of the application has a certain degree of edge intelligence, so that it can make local adjustments under the task framework formulated by the decision agent based on the dynamically changing environment and system state, thereby realizing safer, more efficient, and more accurate inspection.

[0032] Among them, for meteorological environment information, the real-time API of the meteorological bureau or the local micro weather station can be accessed to obtain wind speed / wind direction, precipitation / humidity, light intensity, temperature, etc. Among them, the wind speed / wind direction directly affects the stability, safety and endurance of the unmanned aerial vehicle inspection. Precipitation / humidity affects the working effect of the sensor (especially the infrared thermal imager) and the device surface heat dissipation condition. Light intensity affects the quality of the visible light image, such as overexposure or shadow. The ambient temperature is an important reference benchmark for infrared temperature measurement analysis.

[0033] Among them, after obtaining the fine-tuning logic, the routine cruise mode task is fine-tuned: When the routine inspection agent detects that the load rate of the target line or the target main transformer in the power system exceeds the preset load threshold, the inspection priority of the target line or the target main transformer is raised; When the routine inspection agent determines that the target power equipment is in a power-off maintenance state through the switch state, the target power equipment is automatically skipped in the current routine cruise mode task, and the state is fed back to the decision agent; When the routine inspection agent detects that the meteorological environment corresponding to the current inspection area is lower than the first environment threshold and the regional line load exceeds the preset load area, the inspection route is adjusted; When the routine inspection agent detects that the meteorological environment corresponding to the current inspection area is lower than the second environment threshold, the inspection is stopped and the environment information is fed back to the decision agent; the first environment threshold is higher than the second environment threshold.

[0034] In the embodiment of the application, after the routine cruise task is executed, inspection data is obtained, and the inspection data at least includes visible light images and infrared temperature data of the power equipment. Among them, the real-time initial analysis result includes abnormal power equipment identification, abnormal type and confidence.

[0035] Specifically, the routine inspection agent inspects the power equipment, obtains inspection data, and performs real-time initial analysis according to the inspection data, specifically including: The routine inspection agent synchronously collects visible light images and infrared temperature data of the power equipment through the visible light camera and the infrared thermal imager carried during the cruise process; The routine inspection agent performs real-time initial analysis through the lightweight neural network model deployed locally, to identify meter reading, mechanical deformation and foreign matter attachment from the visible light images, and to identify temperature abnormal points from the infrared temperature data; wherein the real-time initial analysis result is a structured data object, at least including device ID, abnormal type, abnormal position coordinates, abnormal confidence and time stamp. In the embodiment of the application, the lightweight neural network model is pre-trained according to the power equipment abnormal training sample.

[0036] Step 130: The decision-making agent performs comprehensive analysis according to the real-time initial analysis results sent by each regular inspection agent, the historical operation and maintenance records corresponding to the abnormal power equipment, the current power grid operation condition, and the power equipment knowledge base information, to obtain target analysis results, and determines a detailed inspection mode task according to the target analysis results.

[0037] Among them, before obtaining the detailed inspection mode task, model training and setting are first performed in the decision-making agent to enable the decision-making agent to accurately analyze the detailed task. Specifically, the following methods are included: Step 001: Generate detailed inspection training data according to the information of each power equipment of the power grid system, the distribution information of the power equipment, the historical operation and maintenance data, the counter-operation and maintenance data, and the expert experience operation and maintenance data, and the corresponding detailed inspection labels.

[0038] Step 002: Convert the detailed inspection training data into context prompt data, and fine-tune the large model using LoRA to obtain a fine-tuned large model.

[0039] Among them, the decision-making agent constructs a structured decision context prompt by combining the information of each power equipment of the power grid system, the distribution information of the power equipment, the historical operation and maintenance data, the counter-operation and maintenance data, and the expert experience operation and maintenance data, and the corresponding detailed inspection labels, inputs the structured decision context prompt into the large model, and fine-tunes the large model using LoRA to obtain a fine-tuned large model. The fine-tuning of the large model using LoRA can be performed in the existing manner, and the embodiments of the present application do not make specific limitations. The large language model performs semantic understanding and information association, causal reasoning and impact assessment, and decision generation and task planning. Among them, semantic understanding and information association is to deeply understand the multi-source information in the prompt, and to associate the current anomaly with the historical defect record and the equipment family defect knowledge. Causal reasoning and impact assessment is to reason the possible root cause of the current anomaly based on the knowledge of the power system, and to evaluate the potential impact chain on the safety of the equipment and the reliability of the regional power supply. Decision generation and task planning is to generate decision suggestions based on the above reasoning, and the decision suggestions include: triggering detailed inspection immediately, increasing the frequency of regular inspection, or suggesting to combine with the next planned power-off maintenance; when it is suggested to trigger detailed inspection immediately, the guidance content of the detailed inspection mode task is generated at the same time, including the type of detailed inspection sensor to be used and the key inspection parts.

[0040] Step 003: Input the detailed inspection training data into a lightweight neural network for training to obtain a detailed inspection prediction model.

[0041] The lightweight neural network is a lightweight machine learning model, and gradient boosting decision tree (GBDT) can be used in the embodiment of the application to focus on fast and accurate causal inference and prediction. Gradient boosting decision tree (GBDT) can mine complex nonlinear relationships from massive historical data and give quantitative failure probability and risk score. This statistical-based prediction is very accurate and fast. Its decision is completely based on data, avoiding human bias, and can find weak signs and correlation patterns that the human eye cannot detect. And focus on the local, it focuses on the current abnormal equipment itself and its directly related data, and performs deep "focal analysis", providing a solid quantitative basis for decision-making.

[0042] In the embodiment of the application, feature engineering is constructed on the detailed investigation training data to obtain feature data, and the feature data is input into gradient boosting decision tree (GBDT) for iterative training, so as to obtain a trained detailed investigation prediction model.

[0043] Step 004: generating a dynamic abnormal knowledge graph according to the information of each power equipment of the power grid system, the power equipment distribution information, the historical operation and maintenance data, the counter operation and maintenance data and the expert experience operation and maintenance data.

[0044] The dynamic abnormal knowledge graph includes the topological relationship between the power equipment, the historical abnormal record and the defect evolution law. It has the characteristics of explainable and structured causal correlation. The knowledge graph explicitly stores the causal rules such as "A abnormality may cause B defect". Its decision-making process is a white box, which can directly tell the operation and maintenance personnel "why" to handle it in this way, easy to understand and trust. It has the ability of correlation mining, through graph query and reasoning, it can quickly find other potential defects that are strongly associated with the current abnormality, recommend "combined" inspection to avoid missed detection. Moreover, it structures the expert experience, the typical defects of the equipment family and the like, so that the expert knowledge can be inherited and applied on a large scale.

[0045] In the embodiment of the application, first, the graph design is constructed to determine the node type and the relationship type. The node type includes power equipment, defect type, abnormal phenomenon, operation and maintenance measure and the like. The relationship type includes: has_fault (the equipment has a defect), causes (defect A causes phenomenon B), requires_check (defect A needs to check C) and the like. Then, the initial knowledge such as the information of each power equipment of the power grid system, the power equipment distribution information, the historical operation and maintenance data, the counter operation and maintenance data and the expert experience operation and maintenance data is parsed and structured and input into the automatic rule engine to extract the triple information, which is (power equipment, defect, phenomenon). Finally, the graph query reasoning rule is set to obtain the dynamic abnormal knowledge graph. For example, when the abnormality of "transformer bushing temperature is too high" is input, the Cypher query is executed to obtain the corresponding query result.

[0046] The embodiment of the application forms a collaborative, verification, and iterative intelligent decision network through the three models. The strategy can be qualitatively formulated through fine-tuning the large model, the strategy can be quantitatively and quickly formulated through the detailed investigation prediction model, and the correlation and explanation can be performed through the dynamic abnormal knowledge graph.

[0047] After obtaining the three models, according to the real-time initial analysis result sent by each conventional inspection intelligent agent, according to the real-time initial analysis result, the historical operation and maintenance record corresponding to the abnormal power equipment, the current power grid operation condition, and the power equipment knowledge base information, a target analysis result is obtained through comprehensive analysis, and a detailed investigation and inspection mode task is determined according to the target analysis result. Specifically: The real-time initial analysis result, the historical operation and maintenance record corresponding to the abnormal power equipment, and the current power grid operation condition are input into the detailed investigation prediction model to obtain a first detailed investigation task.

[0048] The power equipment information, power equipment distribution information, historical operation and maintenance data, counter-operation and maintenance data, and expert experience operation and maintenance data of the power grid system are input into the fine-tuned large model to obtain a second detailed investigation task. The real-time initial analysis result, the historical operation and maintenance record corresponding to the power equipment, the current power grid operation condition, and the power equipment knowledge base information are collectively constructed into a structured decision context prompt, and then input into the fine-tuned large model to obtain the second detailed investigation task.

[0049] The real-time initial analysis result and the current power grid operation condition are input into the dynamic abnormal knowledge graph to obtain a third detailed investigation task.

[0050] The detailed investigation and inspection mode task is determined according to the first detailed investigation task, the second detailed investigation task, and the third detailed investigation task.

[0051] In the embodiment of the application, the central decision fusion device is set to process the first detailed investigation task, the second detailed investigation task, and the third detailed investigation task. First, the task vectorization is performed to convert the tasks into semantic vectors, and then the similar tasks are merged through similarity clustering. For the tasks in conflict in the first detailed investigation task, the second detailed investigation task, and the third detailed investigation task, corresponding rules are set to eliminate the conflicts. Specifically, the rules can be: Rule 1: If the prediction probability of the detailed investigation prediction model is greater than 0.9, regardless of the results of other models, the task priority is set to the highest.

[0052] Rule 2: If the third detailed investigation task suggested in the fine-tuned large model is found to have strong correlation support in the knowledge graph, the priority of the task is improved.

[0053] Rule 3: If the fine-tuned large model is opposite to the conclusion of the detailed prediction model, the quantitative result of the detailed prediction model is the main basis, but the dissent of the fine-tuned large model is recorded.

[0054] In the embodiment of the application, the priority of each detail task is further ranked according to the weighted scoring formula, and specifically, the score of each detail task is calculated by the following formula:

[0055] For the same detailed task, is the weight of the confidence of the detailed prediction model, is the weight of the correlation degree in the knowledge graph correlation degree, is the weight of the large model importance.

[0056] Through the above score ranking, the final detailed inspection mode task is obtained.

[0057] Step 140: The detailed inspection agent performs detailed inspection and initial detailed analysis on the detailed data according to the detailed inspection mode task, and sends the initial detailed analysis to the decision agent.

[0058] The detailed inspection agent corresponding to the region performs detailed inspection according to the detailed inspection mode task. Specifically, it can include: the detailed ground robot deployed nearby receives the task and autonomously navigates to a certain power equipment, such as under the pole tower #1011, according to the task requirement, raises the mast, activates the partial discharge detector and ultraviolet imager, and performs fine scanning on the specified insulator string to obtain detailed data.

[0059] In the embodiment of the application, the detailed inspection agent also has a lightweight neural network model deployed locally, so that the detailed data can be analyzed in real time. In the embodiment of the application, the lightweight neural network model is trained in advance according to the power equipment abnormal training sample.

[0060] Step 150: The decision agent analyzes the detailed data and the initial detailed analysis to obtain a target detailed analysis result, and performs early warning according to the target detailed analysis result.

[0061] In the embodiment of the application, after obtaining the target detailed analysis result, the decision agent compares the target detailed analysis result with the preset multi-level early warning threshold. According to the comparison result, different levels of early warning response are triggered.

[0062] The different levels of early warning response at least include a first level of early warning, a second level of early warning and a third level of early warning.

[0063] Specifically, when the first level warning is generated, a maintenance work order is generated and included in the planned maintenance process; when the second level warning is generated, an alarm information is sent to the regional operation personnel and the feedback is required within the specified time; when the third level warning is generated, the alarm information is sent to the regional operation personnel, the emergency alarm is sent to the dispatch center and the control system, and the emergency operation strategy is recommended, and the corresponding power equipment is automatically linked and isolated.

[0064] In the above manner, the switching between the two modes can be effectively realized, and the embodiment of the application can effectively improve the inspection efficiency, accuracy, system intelligence and robustness.

[0065] The method of the embodiment of the application is based on multiple power inspection agents, and the power inspection agents include a decision agent, a regular inspection agent and a detailed inspection agent. The decision agent issues an inspection task, and the inspection task includes a regular cruise mode task. The regular inspection agent fine-tunes the inspection task according to the regular cruise mode task, a real-time operation mode of a power grid and meteorological environment information, inspects each power equipment in the power system according to the fine-tuned inspection task, obtains inspection data, performs real-time initial analysis according to the inspection data, and sends the real-time initial analysis result to the decision agent in real time. The decision agent performs comprehensive analysis according to the real-time initial analysis result sent by each regular inspection agent, the real-time initial analysis result, historical operation and maintenance records of the abnormal power equipment, a current power grid operation condition and power equipment knowledge base information, obtains a target analysis result, and determines a detailed inspection mode task according to the target analysis result. The detailed inspection agent performs detailed inspection according to the detailed inspection mode task to the target power equipment, performs initial detailed analysis on the detailed inspection data, and sends the initial detailed analysis to the decision agent. The decision agent performs analysis on the detailed inspection data and the initial detailed analysis, obtains a target detailed analysis result, and performs warning according to the target detailed analysis result. Through the dual-mode linkage control of the power equipment intelligent inspection, the dual-mode intelligent linkage of the regular inspection and the detailed inspection can be realized, and comprehensive analysis and adaptive scheduling can be performed.

[0066] Figure 3 The structure of the dual-mode linkage control system of the power equipment intelligent inspection provided by the embodiment of the application is shown. As shown in Figure 3 The system 300 includes multiple power inspection agents, and the power inspection agents include a decision agent 310, a regular inspection agent 320 and a detailed inspection agent 330.

[0067] The decision agent 310 is configured to issue an inspection task, and the inspection task includes a regular cruise mode task. The regular cruise mode task includes a regular route, an inspection frequency and an inspection content of the regular cruise mode task. The conventional inspection agent 320 is used to fine-tune the inspection task according to the conventional patrol mode task, the real-time operation mode of the power grid, and meteorological environment information, and to inspect each power equipment in the power system according to the fine-tuned inspection task, acquire inspection data, perform real-time initial analysis based on the inspection data, and send the real-time initial analysis results to the decision agent 310 in real time; the real-time initial analysis results include abnormal power equipment identification, abnormality type, and confidence level; The decision-making intelligent agent 310 is also used to perform comprehensive analysis based on the real-time initial analysis results sent by each regular inspection intelligent agent 320, the historical operation and maintenance records of the corresponding abnormal power equipment, the current power grid operation status, and the power equipment knowledge base information to obtain the target analysis result, and determine the detailed inspection mode task based on the target analysis result. The detailed inspection agent 330 is used to conduct detailed inspections at the target power equipment according to the detailed inspection mode task, and after performing initial detailed inspection analysis on the detailed inspection data, send it to the decision agent 310. The decision-making agent 310 is further configured to obtain the target detailed investigation analysis result after analyzing the detailed investigation data and the initial detailed investigation analysis, and to issue an early warning based on the target detailed investigation analysis result.

[0068] In an alternative embodiment, the decision-making agent 310 is further configured to: Information on each power device and its distribution in the power system is obtained from the power system's equipment information database. The power device information includes the type of power device, physical parameters, and electrical parameters. The power device distribution information includes the geographical location of each power device and the electrical connection relationships between them. By inputting the information of each power device and the distribution information of the power devices into the finely tuned large model, the conventional cruise mode task is obtained.

[0069] In an alternative embodiment, the decision-making agent 310 is further configured to: Based on the information of each power equipment in the power grid system, the distribution information of power equipment, historical operation and maintenance data, countermeasure operation and maintenance data, and expert experience operation and maintenance data, as well as the corresponding detailed inspection and patrol tags, detailed inspection training data is generated. The detailed training data is converted into contextual cue data, and LoRA is used to fine-tune the large model to obtain a fine-tuned large model. The detailed investigation training data is input into a lightweight neural network for training to obtain a detailed investigation prediction model; According to the power equipment information, power equipment distribution information, historical operation and maintenance data, counter operation and maintenance data and expert experience operation and maintenance data of the power grid system, a dynamic abnormal knowledge graph is generated; the dynamic abnormal knowledge graph includes the topological relationship between power equipment, historical abnormal records and defect evolution law.

[0070] In an optional manner, the decision agent 310 is further configured to: input the real-time initial analysis result, the historical operation and maintenance record corresponding to the abnormal power equipment and the current power grid operation condition into a detailed investigation prediction model to obtain a first detailed investigation task; input the power equipment information, power equipment distribution information, historical operation and maintenance data, counter operation and maintenance data and expert experience operation and maintenance data of the power grid system into the fine-tuned large model to obtain a second detailed investigation task; input the real-time initial analysis result and the current power grid operation condition into the dynamic abnormal knowledge graph to obtain a third detailed investigation task; determine the detailed investigation and inspection mode task according to the first detailed investigation task, the second detailed investigation task and the third detailed investigation task.

[0071] In an optional manner, the regular inspection agent 320 is further configured to: when the regular inspection agent 320 detects that the load rate of a target line or a target main transformer in the power system exceeds a preset load threshold, the inspection priority of the target line or the target main transformer is increased; when the regular inspection agent 320 determines that a target power equipment is in a power-off maintenance state through the switch state, the target power equipment is automatically skipped in the current regular cruise mode task, and a state feedback is given to the decision agent 310; when the regular inspection agent 320 detects that the meteorological environment corresponding to the current inspection area is lower than a first environment threshold and the load of the regional line exceeds a preset load area, the inspection route is adjusted; when the regular inspection agent 320 detects that the meteorological environment corresponding to the current inspection area is lower than a second environment threshold, the inspection is stopped and the environment information is fed back to the decision agent 310; the first environment threshold is higher than the second environment threshold.

[0072] In an optional manner, the decision agent 310 performs early warning according to the target detailed investigation and analysis result, and specifically includes: the decision agent 310 compares the target detailed investigation and analysis result with a plurality of preset early warning thresholds; according to the comparison result, different levels of early warning responses are triggered; the different levels of early warning responses at least include a first level early warning, a second level early warning and a third level early warning; When the first level of early warning is given, a maintenance work order is generated and included in the planned maintenance process; When the second level of early warning is given, an alarm message is sent to the regional operation and maintenance personnel, and feedback is required within a specified time; When the third level of early warning is given, an alarm message is sent to the regional operation and maintenance personnel, an emergency alarm is sent to the dispatch center and the control system, and an emergency operation strategy is recommended, and the corresponding power equipment is automatically isolated.

[0073] In an optional manner, the inspection data at least includes visible light images and infrared temperature data of the power equipment; the routine inspection agent 320 inspects the power equipment, acquires the inspection data, and performs real-time initial analysis according to the inspection data, specifically including: The routine inspection agent 320 synchronously collects visible light images and infrared temperature data of the power equipment through the visible light camera and the infrared thermal imager carried during the cruising process; The routine inspection agent 320 performs real-time initial analysis through the lightweight neural network model deployed locally, to identify meter reading, mechanical deformation and foreign object attachment from the visible light images, and to identify temperature abnormal points from the infrared temperature data; wherein the real-time initial analysis result is a structured data object, at least including equipment ID, abnormal type, abnormal position coordinates, abnormal confidence and time stamp.

[0074] The method of the embodiment of the application is based on a plurality of power inspection intelligent agents, the power inspection intelligent agents including a decision intelligent agent, a regular inspection intelligent agent and a detailed inspection intelligent agent. The decision intelligent agent issues an inspection task; the inspection task includes a regular cruise mode task; the regular inspection intelligent agent fine-tunes the inspection task according to the regular cruise mode task, a real-time operation mode of a power grid and meteorological environment information, and inspects each power equipment in the power system according to the fine-tuned inspection task, acquires inspection data, performs real-time initial analysis according to the inspection data, and sends the real-time initial analysis result to the decision intelligent agent in real time; the decision intelligent agent performs comprehensive analysis according to the real-time initial analysis result sent by each regular inspection intelligent agent, the real-time initial analysis result, historical operation and maintenance records of the abnormal power equipment, current power grid operation conditions and power equipment knowledge base information, obtains a target analysis result, and determines a detailed inspection mode task according to the target analysis result; the detailed inspection intelligent agent performs detailed inspection to the target power equipment according to the detailed inspection mode task, and sends the initial detailed analysis result to the decision intelligent agent after performing initial detailed analysis on the detailed inspection data; the decision intelligent agent obtains a target detailed analysis result after performing analysis on the detailed inspection data and the initial detailed analysis result, and performs early warning according to the target detailed analysis result. Through the dual-mode linkage control of the intelligent power equipment inspection, the dual-mode intelligent linkage of the regular inspection and the detailed inspection can be realized, and comprehensive analysis and adaptive scheduling can be performed.

[0075] The algorithms and displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general purpose systems can be used with programs in accordance with the teachings herein, or it can prove convenient to construct more specialized apparatus to perform the required method steps. The required structure for a variety of these systems will be apparent from the description above. In addition, the present embodiments are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages can be used to implement the teachings of the present embodiments as described herein, and any references below to specific languages are provided for disclosure of enablement of the best mode of the invention.

[0076] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been described in detail in order to avoid obscuring the understanding of this description.

[0077] Similarly, it is to be understood that the embodiments of the application can be adapted to other applications and environments without departing from the spirit and scope of the application. Also, it is to be understood that specific features of the embodiments of the application can sometimes be interchanged and / or substituted for each other without departing from the spirit and scope of the application.

[0078] Those skilled in the art will understand that modules in the apparatus in the embodiments can be adapted and placed in one or more apparatuses other than the embodiments. Modules or units or components in the embodiments can be combined into one module or unit or component, and can be divided into multiple sub-modules or sub-units or sub-components. All features disclosed in the specification (including the claims, abstract, and drawings) and any method or process or apparatus so disclosed can be combined in any combination, except where such features or steps are mutually exclusive. Each feature disclosed in the specification (including the claims, abstract, and drawings) can be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

[0079] It should be noted that the above-mentioned embodiments illustrate rather than limit the application, and that one skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word 'comprising' does not exclude the presence of elements or steps other than those listed in a claim. The word 'a' or 'an' preceding an element does not exclude the presence of a plurality of such elements. The application can be implemented by means of both hardware and software, and any combination thereof. In a unit claim, several devices can be listed, comprising means for performing the recited steps. Means can be implemented by one and the same item of hardware. The word 'first','second', 'third', etc. do not imply any order. The terms 'comprise', 'comprising', 'include', 'including', and 'includes' should be construed as non-limiting. The steps of the methods disclosed in the specification, except where expressly stated otherwise, are not to be construed as necessarily requiring the steps to be performed in the order disclosed.

Claims

1. A dual-mode linkage control method for intelligent inspection of power equipment, characterized in that, The method is based on multiple power line inspection agents, including decision-making agents, routine inspection agents, and detailed inspection agents; the method includes: Inspection tasks are issued through a decision-making intelligent agent; the inspection tasks include regular patrol mode tasks; the regular patrol mode tasks include the regular routes, inspection frequencies and inspection contents of the regular patrol mode tasks. The conventional inspection agent fine-tunes the inspection task based on the conventional patrol mode, the real-time operation mode of the power grid, and meteorological information. It then inspects each power device in the power system according to the fine-tuned inspection task, acquires inspection data, performs real-time initial analysis based on the inspection data, and sends the real-time initial analysis results to the decision agent in real time. The real-time initial analysis results include abnormal power device identification, abnormality type, and confidence level. The decision-making intelligent agent performs a comprehensive analysis based on the real-time initial analysis results sent by each regular inspection intelligent agent, the historical operation and maintenance records of the corresponding abnormal power equipment, the current power grid operating conditions, and the power equipment knowledge base information to obtain the target analysis result, and determines the detailed inspection mode task based on the target analysis result. The detailed inspection agent goes to the target power equipment to conduct a detailed inspection according to the detailed inspection mode task, and after performing an initial detailed inspection analysis on the detailed inspection data, sends the initial detailed inspection analysis to the decision-making agent. The decision-making agent analyzes the detailed investigation data and the initial detailed investigation analysis to obtain the target detailed investigation analysis results, and issues an early warning based on the target detailed investigation analysis results.

2. The method according to claim 1, characterized in that, Before issuing the inspection task through the decision-making intelligent agent, the method further includes: The decision-making agent obtains information on each power device and the distribution information of the power devices in the power system based on the power system's equipment information database. The power device information includes the type of power device, physical parameters, and electrical parameters. The power device distribution information includes the geographical location of each power device and the electrical connection relationships between each power device. The decision-making agent inputs information about each power device and the distribution of power devices into the finely tuned large model to obtain the conventional cruise mode task.

3. The method according to claim 1, characterized in that, The decision-making agent, based on the real-time initial analysis results sent by each regular inspection agent, and considering the real-time initial analysis results, the historical operation and maintenance records of the corresponding abnormal power equipment, the current power grid operating conditions, and power equipment knowledge base information, performs a comprehensive analysis to obtain the target analysis result. Before determining the detailed inspection mode task based on the target analysis result, the method includes: Based on the information of each power equipment in the power grid system, the distribution information of power equipment, historical operation and maintenance data, countermeasure operation and maintenance data, and expert experience operation and maintenance data, as well as the corresponding detailed inspection and patrol tags, detailed inspection training data is generated. The detailed training data is converted into contextual cue data, and LoRA is used to fine-tune the large model to obtain a fine-tuned large model. The detailed investigation training data is input into a lightweight neural network for training to obtain a detailed investigation prediction model; A dynamic anomaly knowledge graph is generated based on information about each power device in the power grid system, power device distribution information, historical operation and maintenance data, countermeasure operation and maintenance data, and expert experience operation and maintenance data. The dynamic anomaly knowledge graph includes the topological relationships between power devices, historical anomaly records, and defect evolution patterns.

4. The method according to claim 3, characterized in that, The decision-making agent, based on the real-time initial analysis results sent by each regular inspection agent, and considering the real-time initial analysis results, the historical operation and maintenance records of the corresponding abnormal power equipment, the current power grid operating conditions, and power equipment knowledge base information, performs a comprehensive analysis to obtain the target analysis result. Based on the target analysis result, it determines the detailed inspection mode task, including: The real-time initial analysis results, the historical operation and maintenance records of the corresponding abnormal power equipment, and the current power grid operating conditions are input into the detailed investigation and prediction model to obtain the first detailed investigation task; The information of each power equipment in the power grid system, the distribution information of the power equipment, the historical operation and maintenance data, the countermeasure operation and maintenance data, and the expert experience operation and maintenance data are input into the fine-tuned large model to obtain the second detailed investigation task; The real-time initial analysis results and the current power grid operating conditions are input into the dynamic anomaly knowledge graph to obtain the third detailed investigation task; The detailed inspection mode task is determined based on the first detailed inspection task, the second detailed inspection task, and the third detailed inspection task.

5. The method according to any one of claims 1-4, characterized in that, The conventional inspection agent fine-tunes the inspection task based on the conventional patrol mode, the real-time operation mode of the power grid, and meteorological information. It then inspects various power devices in the power system according to the adjusted task, acquires inspection data, performs real-time initial analysis based on the inspection data, and sends the results of the real-time initial analysis to the decision-making agent in real time. This includes: When the conventional inspection intelligent agent detects that the load rate of the target line or the target main transformer in the power system exceeds the preset load threshold, the inspection priority of the target line or the target main transformer is increased. When the routine inspection agent determines that the target power equipment is in a power outage maintenance state based on the switch status, it automatically skips the target power equipment in the current routine patrol mode task and feeds back the status to the decision-making agent. When the conventional inspection agent detects that the meteorological environment of the current inspection area is lower than the first environmental threshold and the line load of the area exceeds the preset load area, the inspection route is adjusted. When the routine inspection agent detects that the meteorological environment of the current inspection area is lower than the second environmental threshold, it stops the inspection and feeds back the environmental information to the decision-making agent; the first environmental threshold is higher than the second environmental threshold.

6. The method according to any one of claims 1-4, characterized in that, The decision-making agent issues early warnings based on the detailed analysis results of the target, specifically including: The decision-making agent compares the results of the detailed investigation and analysis of the target with the preset multi-level early warning thresholds; Based on the comparison results, different levels of early warning responses are triggered; the different levels of early warning responses include at least Level 1, Level 2, and Level 3 early warnings. When the warning level is 1, a maintenance work order is generated and incorporated into the planned maintenance process. When a Level 2 warning is issued, an alarm message is sent to the regional operations and maintenance personnel, who are required to provide feedback within a specified time. When a Level 3 warning is issued, an alarm message is sent to regional maintenance personnel, and an emergency alarm is sent to the dispatch center and control system. Emergency operation strategies are recommended, and the corresponding power equipment is automatically isolated.

7. The method according to any one of claims 1-4, characterized in that, The inspection data includes at least visible light images and infrared temperature data of the power equipment; The conventional inspection intelligent agent inspects power equipment, acquires inspection data, and performs real-time initial analysis based on the inspection data, specifically including: During the patrol, the conventional inspection intelligent agent simultaneously collects visible light images and infrared temperature data of the power equipment using a visible light camera and an infrared thermal imager. The conventional inspection agent performs real-time initial analysis using a locally deployed lightweight neural network model to identify meter readings, mechanical deformation, and foreign object attachment from the visible light image, and to identify temperature anomalies from the infrared temperature data. The real-time initial analysis results are structured data objects, including at least device ID, anomaly type, anomaly location coordinates, anomaly confidence level, and timestamp.

8. A dual-mode linkage control system for intelligent inspection of power equipment, characterized in that, The system includes multiple power line inspection intelligent agents, including decision-making intelligent agents, routine inspection intelligent agents, and detailed inspection intelligent agents; wherein: The decision-making agent is used to issue inspection tasks; the inspection tasks include regular patrol mode tasks; the regular patrol mode tasks include the regular route, inspection frequency and inspection content of the regular patrol mode tasks. The conventional inspection agent is used to fine-tune the inspection task according to the conventional patrol mode task, the real-time operation mode of the power grid, and meteorological environment information, and to inspect each power equipment in the power system according to the fine-tuned inspection task, acquire inspection data, perform real-time initial analysis based on the inspection data, and send the real-time initial analysis results to the decision agent in real time; the real-time initial analysis results include abnormal power equipment identification, abnormality type, and confidence level; The decision-making intelligent agent is also used to perform comprehensive analysis based on the real-time initial analysis results sent by each regular inspection intelligent agent, the historical operation and maintenance records of the corresponding abnormal power equipment, the current power grid operating conditions, and the power equipment knowledge base information to obtain the target analysis result, and determine the detailed inspection mode task based on the target analysis result. The detailed inspection agent is used to conduct detailed inspections at the target power equipment according to the detailed inspection mode task, and after performing initial detailed inspection analysis on the detailed inspection data, send it to the decision-making agent. The decision-making agent is also used to obtain the target detailed investigation analysis result after analyzing the detailed investigation data and the initial detailed investigation analysis, and to issue an early warning based on the target detailed investigation analysis result.

9. The system according to claim 8, characterized in that, The decision-making agent is also used for: Based on the information of each power equipment in the power grid system, the distribution information of power equipment, historical operation and maintenance data, countermeasure operation and maintenance data, and expert experience operation and maintenance data, as well as the corresponding detailed inspection and patrol tags, detailed inspection training data is generated. The detailed training data is converted into contextual cue data, and LoRA is used to fine-tune the large model to obtain a fine-tuned large model. The detailed investigation training data is input into a lightweight neural network for training to obtain a detailed investigation prediction model; A dynamic anomaly knowledge graph is generated based on information about each power device in the power grid system, power device distribution information, historical operation and maintenance data, countermeasure operation and maintenance data, and expert experience operation and maintenance data. The dynamic anomaly knowledge graph includes the topological relationships between power devices, historical anomaly records, and defect evolution patterns.

10. The system according to claim 9, characterized in that, The decision-making agent is also used for: The real-time initial analysis results, the historical operation and maintenance records of the corresponding abnormal power equipment, and the current power grid operating conditions are input into the detailed investigation and prediction model to obtain the first detailed investigation task; The information of each power equipment in the power grid system, the distribution information of the power equipment, the historical operation and maintenance data, the countermeasure operation and maintenance data, and the expert experience operation and maintenance data are input into the fine-tuned large model to obtain the second detailed investigation task; The real-time initial analysis results and the current power grid operating conditions are input into the dynamic anomaly knowledge graph to obtain the third detailed investigation task; The detailed inspection mode task is determined based on the first detailed inspection task, the second detailed inspection task, and the third detailed inspection task.

Images