A direct current control and protection auxiliary decision method and system based on multi-agent cooperation technology

Abstract

The application provides a direct current control and protection auxiliary decision-making method and system based on multi-agent cooperation technology, belongs to the technical field of direct current control and protection, and comprises a data processing agent, a protection action analysis agent, a fault positioning agent, a fault disposal scheme generation agent and a global cooperative scheduling agent, and is deployed with three-level standardized skill units of basic general skills, special business skills and cooperative scheduling skills; the global cooperative scheduling agent is configured to overall plan the system operation, and dynamically issue algorithm operation parameters to each business agent and skill unit. The advantage lies in that a weighted fusion positioning system of four dimensions of model reasoning, electrical characteristics, protection logic and regional topology is constructed, a multi-source heterogeneous score is weighted and summed through a dynamic weight coefficient adaptively distributed by the global cooperative scheduling agent, and the positioning precision problem caused by single electrical quantity or simple logic matching of an existing scheme is overcome.

Description

Technical Field

[0001] This invention belongs to the field of DC control and protection technology, specifically relating to a DC control and protection auxiliary decision-making method and system based on multi-agent cooperative technology. Background Technology

[0002] In recent years, the state has continuously promoted the construction of new power systems and the intelligent upgrading of energy equipment, providing clear policy guidance for technological innovation in the field of DC control and protection. Accelerating the development of key equipment such as high-reliability, large-capacity, low-loss DC converter transformers, converter valves based on high-parameter power devices, and ultra-high-voltage DC control and protection systems will improve the efficiency and stability of long-distance, high-power power transmission. Simultaneously, the development of AI-based automated control systems and the creation of intelligent grid control software with autonomous decision-making capabilities will promote the upgrading of the grid's intelligent regulation and control system. Concurrently, focusing on grid security and new energy consumption requirements under the new power system, the application of AI models in intelligent auxiliary decision-making and dispatch control of the grid will be explored, creating intelligent agents for power equipment health management with autonomous perception, decision-making, and execution capabilities. In high-voltage / ultra-high-voltage direct current (HVDC) transmission systems, DC control and protection (DC protection) equipment is a core component ensuring the safe and stable operation of the power grid. Due to the complex structure and variable operating conditions of DC systems, failures in DC protection equipment often involve multi-station coordination and multi-source data linkage, resulting in a highly nonlinear correlation between fault phenomena and their causes. Furthermore, the multi-modal data of DC systems encompasses heterogeneous modes such as time-series electrical quantities, transient waveform data, and equipment condition monitoring data, with significant differences in data structure, sampling frequency, and semantic representation. Direct fusion can easily lead to a disordered feature space. Against this backdrop, constructing a DC control and protection auxiliary decision-making system adapted to complex operating conditions has become a key technical issue for implementing the national intelligent power strategy and supporting the construction of new power systems.

[0003] In high-voltage / ultra-high-voltage direct current (HVDC) transmission systems, DC control and protection equipment is a core component ensuring the safe and stable operation of the power grid. Due to the complex structure and variable operating conditions of DC systems, failures in control and protection equipment often involve multi-station coordination and multi-source data linkage, resulting in a highly nonlinear correlation between fault phenomena and their causes. Currently, while DC control and protection auxiliary decision-making technology has achieved a certain research foundation, significant industry-specific shortcomings remain in five key areas: protection action verification, accurate fault location, operation and maintenance scheme generation, cross-site generalization and adaptation, and system autonomous iteration. These shortcomings make it difficult to meet the needs of accurate decision-making under complex operating conditions.

[0004] Existing DC control and protection auxiliary decision-making schemes have the disadvantages of poor model adaptability, the inability of a single model or fixed parameter fusion strategy to adapt to the differentiated characteristics of different converter stations and different fault scenarios, and weak cross-station generalization ability. Summary of the Invention

[0005] In view of the above, this invention addresses the shortcomings of existing technologies by providing a DC control and protection auxiliary decision-making method and system based on multi-agent cooperative technology. To solve the aforementioned technical problems, the technical solution adopted by this invention includes: S1. The data processing agent executes the basic general skill, performs adaptive standardization preprocessing on the collected multi-source heterogeneous fault data based on the corresponding algorithm running parameters, and pushes the processed standardized data to the protection action analysis agent according to the time sequence logic. S2, the protection action analysis agent executes a special business skill, a bidirectional iterative fusion algorithm of large and small models, performs joint analysis and source verification of protection actions based on the corresponding algorithm running parameters, and outputs a dual-model fusion judgment result containing model inference parameters; S3. The fault location agent executes a special business Skill, calls a multi-feature weighted fusion location algorithm, and calculates the weighted sum of the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score based on the corresponding algorithm running parameters to obtain the comprehensive fault matching score. S4. The fault handling solution generation intelligent agent executes the special business Skill, constructs multi-dimensional search tags based on the model inference parameters, fault type, fault location, and electrical characteristics, determines the case matching weight based on the corresponding algorithm running parameters, calls the multi-dimensional similar case matching algorithm to retrieve historical cases, extracts the standard maintenance steps in similar cases, and generates a maintenance plan by combining the fault severity level and operation and maintenance resource configuration data. S5, the global collaborative scheduling intelligent agent summarizes the weighted sum of the error loss values ​​of the four stages of data processing, protection action analysis, fault location and case matching, updates the parameters of the whole system based on the Skill parameter dynamic iterative algorithm, and sends the corresponding algorithm running parameters to each business intelligent agent and Skill unit.

[0006] Furthermore, the basic general skills include event awareness skills, data acquisition skills, data preprocessing skills, and data forwarding skills; Step S1 includes: S1.1 The data processing intelligent agent continuously monitors the alarm signals of the field business system through the event perception Skill, classifies and identifies DC control and protection alarm information, and prioritizes responding to core alarms of equipment abnormality and protection action. S1.2 Establish a multi-threaded parallel acquisition mechanism through the data acquisition skill to synchronously capture DC control and protection fault waveform timing data and SER event text data; S1.3. Start the data preprocessing skill and use the cross-site data adaptive normalization algorithm to standardize the collected raw electrical quantity data. The operation of the cross-site data adaptive normalization algorithm is based on the algorithm operation parameters issued by the global collaborative scheduling agent. S1.4 Data forwarding Skill follows the preset business interaction protocol, pushes the pre-processed standardized dataset to the downstream protection action analysis agent according to the time sequence logic, and transmits the full process error of this stage back to the global collaborative scheduling agent in real time for global iterative calculation.

[0007] Furthermore, step S2 includes: S2.1 The protection action analysis intelligent agent automatically matches a small power mechanism model that is compatible with the current fault scenario based on fault semantic features and operating condition labels through a special business skill. S2.2 Perform bidirectional iterative fusion calculation of the size model, the calculation including: S2.21. Using the output of the large model and the threshold correction weight issued by the global collaborative scheduling agent, the initial baseline threshold of the mechanism small model is dynamically optimized to obtain the optimized protection action judgment threshold. S2.22. Using the physical calculation results of the mechanism small model and the dynamic fusion weights issued by the global collaborative scheduling agent, the semantic reasoning judgment score of the large model is corrected for deviation; the final judgment result after the fusion of the two models is output. S2.23. Based on the final judgment result after the fusion of the two models, generate a structured verification conclusion that includes the protection action triggering sequence, logic matching degree and deviation correction basis to verify the compliance of the protection action; S2.3. Through multi-level thinking chain reasoning skills, the fault analysis logic is broken down layer by layer, and the data and judgment basis of the whole process of model parameter adjustment, rule comparison and result correction are fully recorded to construct a reproducible reasoning trajectory. S2.4. The judgment conclusion of the dual-model fusion output is cross-validated by the result verification Skill, abnormal deviation data generated by single model operation is removed, and the abnormal marker is sent back to the global collaborative scheduling agent. S2.5. Compile and analyze the data, generate a standardized protection action analysis report, and push the core fault characteristics and fusion judgment results to the downstream fault location intelligent agent according to the time sequence logic. At the same time, the calculation error of this link is sent back to the global collaborative scheduling intelligent agent for global iterative calculation.

[0008] Furthermore, step S3 includes: S3.1 The fault location intelligence receives the dual-model fusion analysis conclusion issued by the upstream protection action analysis intelligence. The fault preliminary judgment Skill combines the upstream dual-model fusion analysis conclusion to perform preliminary judgment clustering of fault types and narrow the scope of fault judgment. S3.2 Channel Data Matching Skill: Based on the initial fault type, intelligently filter the core data of three-phase voltage, current, differential current and switching quantity associated with the fault, and eliminate invalid data dimensions. S3.3 Electrical quantity characteristic analysis: The mechanism mini-model driven by Skill extracts the amplitude, sudden change trend, fluctuation period and phase deviation of electrical quantities at the time of fault, and constructs a standardized fault feature vector set. S3.4. Call the multi-feature weighted fusion localization algorithm. Based on the dynamic weight coefficients issued by the global collaborative scheduling agent, perform weighted summation calculation on the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score to obtain the comprehensive fault matching score and lock the specific location and type of the fault. S3.5. Through multi-level thinking chain tracing skill, the entire process of reasoning steps and calculation parameters of fault initial judgment, feature extraction, multi-dimensional fusion analysis and result determination are fully recorded to support the verification and tracing of fault location results. S3.6 Integrate the entire process analysis data and inference logs, unify the data format, and push it to the fault handling solution generation agent. The positioning error of this step is transmitted back to the global collaborative scheduling agent to participate in the global parameter iterative optimization.

[0009] Furthermore, the fault handling scheme generating agent receives standardized location results and inference logs from the fault location agent. The standardized location results include fault type, fault location, and model inference parameters generated by the protection action analysis agent. Step S4 includes: S4.1 Extract multi-dimensional retrieval tags based on the fault type, fault location, electrical characteristics, and model inference parameters generated by the protection action analysis agent, and construct a retrieval vector; call the case similarity matching algorithm, and calculate the similarity between the retrieval vector and the feature vector in the historical fault case library based on the weight coefficients dynamically issued by the global collaborative scheduling agent; S4.2 Based on the similar cases obtained from the search, extract the standard maintenance steps in the case; combine the current fault severity level, the operating status of equipment in the station and the operation and maintenance resource configuration data to generate an inspection plan and maintenance scheme adapted to this fault. S4.3 Push the generated fault analysis report, similar fault cases and customized maintenance plan to the field operation and maintenance terminal to provide data support for front-line fault handling; S4.4 Archive all original data, model parameters, inference logs, and generated repair plans for this fault to the business database, and send the calculation errors in the case matching process back to the global collaborative scheduling agent, and update the historical case library and model training sample set simultaneously to provide data support for the autonomous iterative optimization of the global collaborative scheduling agent.

[0010] Furthermore, step S5 includes: S5.1. By controlling the timing of processes through Skills, the triggering order and data interaction protocol of each intelligent agent and Skill unit are standardized, and the timing constraints for cross-module collaborative execution are established. S5.2 reads the configuration files of each converter station by dynamically adapting the Skill and dynamically iterates the algorithm based on the Skill parameters to adjust the data preprocessing rules, model fusion weights and fault judgment thresholds in order to adapt to the differentiated operating environment of different converter stations. S5.3. Real-time monitoring of the operating status of all system modules through the fault tolerance skill. Automatically trigger retry mechanism, parameter rollback mechanism or abnormal reporting process for data interruption, model call failure or inference abnormal conditions. S5.4 summarizes the full-process error loss value of the four stages of data processing, protection action analysis, fault location and case matching, assigns corresponding weighting coefficients to each component and sums them to obtain the global loss value; based on the Skill parameter dynamic iteration algorithm, the global loss value is used to synchronously update and distribute the algorithm running parameters, all Skill scheduling logic, model operation parameters and judgment weights in the whole system.

[0011] A DC control and protection auxiliary decision-making system based on multi-agent cooperative technology, applied to the DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in any of the preceding claims, characterized in that it includes: The system includes a data processing agent, a protection action analysis agent, a fault location agent, a fault handling plan generation agent, and a global collaborative scheduling agent, and is equipped with three levels of standardized skill units: basic general skills, specialized business skills, and collaborative scheduling skills. The global collaborative scheduling agent is configured to coordinate the operation of the entire system and dynamically distribute algorithm running parameters to each business agent and Skill unit; The data processing agent is used to execute a basic general skill, perform adaptive standardization preprocessing on the collected multi-source heterogeneous fault data based on the corresponding algorithm running parameters, and push the processed standardized data to the protection action analysis agent according to the time sequence logic. The protection action analysis agent is used to execute a special business skill, call the bidirectional iterative fusion algorithm of large and small models, perform joint analysis and source verification of protection actions based on the corresponding algorithm running parameters, and output the dual-model fusion judgment result containing model inference parameters. The fault location intelligent agent is used to execute a special business Skill, call a multi-feature weighted fusion location algorithm, and perform weighted summation calculation on the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score based on the corresponding algorithm running parameters to obtain a comprehensive fault matching score. The fault handling solution generates an intelligent agent to execute a special business skill. It constructs multi-dimensional search tags based on the model inference parameters, fault type, fault location, and electrical characteristics. It determines the case matching weight based on the corresponding algorithm running parameters, calls the multi-dimensional similar case matching algorithm to retrieve historical cases, extracts standard maintenance steps from similar cases, and generates a maintenance plan by combining the fault severity level and operation and maintenance resource configuration data. The global collaborative scheduling agent is used to summarize the weighted sum of error loss values ​​across the four stages of data processing, protection action analysis, fault location, and case matching. Based on the Skill parameter dynamic iteration algorithm, it synchronously and autonomously updates the parameters of the entire system and distributes the corresponding algorithm running parameters to each business agent and Skill unit.

[0012] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the method as described in any one of claims 1-6.

[0013] A computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the method as described in any one of claims 1-6.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention overcomes the problems of insufficient sensitivity of single models or fixed threshold strategies to complex and hidden faults, and the lack of domain knowledge constraints in large models, which easily leads to illusory reasoning, by using a bidirectional iterative fusion mechanism of large and small models. The large model outputs the judgment threshold of the small model to optimize the mechanism, and the physical operation results of the small model correct the semantic reasoning deviation of the large model. This invention can reduce the risk of misjudgment and omission, and reduce the dependence on manual work by highly skilled professional technicians.

[0015] 2. This invention constructs a weighted fusion positioning system based on four dimensions: model reasoning, electrical characteristics, protection logic, and regional topology. By using dynamically weighted coefficients adaptively allocated by a globally coordinated intelligent agent, the system weights and sums the heterogeneous scores from multiple sources, overcoming the problem of insufficient positioning accuracy caused by single electrical quantities or simple logic matching in existing solutions. At the same time, based on the multi-level thinking chain tracing skill, the system fully records the entire reasoning process from initial fault judgment, feature extraction, multi-dimensional fusion analysis, and result determination, enabling the positioning results to have verifiable and traceable technical effects.

[0016] 3. This invention uses the model reasoning parameters generated by the protective action analysis intelligent agent and the fault type, fault location, and electrical characteristics to construct multi-dimensional search tags, retrieve historical similar cases and extract standard maintenance steps, and dynamically generate maintenance plans by combining the current fault severity level, the operating status of equipment in the station and the configuration data of operation and maintenance resources. This overcomes the shortcomings of existing templated plans that are disconnected from on-site constraints and have poor engineering operability, so that the generated inspection plan and maintenance plan can truly adapt to the specific scenario and on-site resource conditions of this fault.

[0017] 4. This invention dynamically distributes algorithm operation parameters (including site adaptive correction coefficients, correction weights, dynamic fusion weights, and multi-feature judgment dynamic weights) to each business intelligent agent and Skill unit through a global collaborative scheduling intelligent agent. This enables the same multi-agent cluster and three-level Skill system to adapt to the differentiated operating environment and fault scenarios of different converter stations. It overcomes the shortcomings of existing solutions, such as the difficulty in directly transferring single-site training models, poor spare parts interoperability, and poor operation and maintenance consistency. This improves the system's cross-site generalization ability and engineering reuse value.

[0018] 5. This invention uses a globally coordinated intelligent agent to uniquely collect and calculate the error loss value of the entire process across four stages. Based on the Skill parameter dynamic iteration algorithm, it synchronously and autonomously updates the parameters of the entire system. It constructs a technical transmission logic from parameter distribution, execution analysis, error feedback, global collection to synchronous iteration. This overcomes the shortcomings of existing systems where each functional module operates independently and lacks a global coordinated scheduling and parameter feedback mechanism, resulting in the system's inability to evolve autonomously. This enables the system to continuously optimize itself during continuous operation, adapting to the complex evolution scenarios of high-proportion renewable energy grid connection and high-power long-distance power transmission under the new power system.

[0019] 6. This invention achieves component decoupling and flexible orchestration of business capabilities through a modular architecture of five types of intelligent agent clusters and three levels of standardized skills. Under global collaborative scheduling, each intelligent agent cooperates, communicates data, and coordinates business, changing the situation where existing auxiliary decision-making schemes rely on manual comprehensive analysis and have delayed response. It can quickly complete the full-link auxiliary decision-making of verification, positioning, and scheme generation after the protection action is triggered, providing technical support for meeting the requirements of new power systems for the safe and stable operation of the power grid. Attached Figure Description

[0020] The present invention will now be described in further detail with reference to the accompanying drawings.

[0021] Figure 1 This is a schematic diagram of the entire operation mechanism of the intelligent agent in this invention; Figure 2 This is a schematic diagram of the three-level Skill modular unit classification and multi-agent coupling distribution in this invention; Figure 3 This is a schematic diagram of the operation process of the innovative algorithm for bidirectional iterative fusion of large and small models in this invention. Detailed Implementation

[0022] To better understand the present invention, the content of the invention is further clearly illustrated below with reference to embodiments and accompanying drawings. However, the scope of protection of the present invention is not limited to the embodiments described below. Numerous specific details are set forth in the following description to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the present invention can be practiced without one or more of these details.

[0023] The existing DC control and protection fault analysis and maintenance model has the following four core technical shortcomings: First, the challenge of adapting heterogeneous data from multiple converter stations: waveform recordings, event messages, and electrical dimensions are not uniform across stations, and general algorithms lack dynamic correction capabilities, resulting in high costs and long development cycles for customized development at each station; Second, the challenge of accuracy in judging complex DC faults: in scenarios involving multiple faults, high-resistance grounding, and commutation failures, single-mechanism models have poor generalization, and pure large models lack electrical constraints, leading to frequent misjudgments and missed judgments; Third, the challenge of fragmented multi-business modules and inability to achieve global collaborative optimization: data processing, action verification, fault location, and solution generation operate independently, and parameters cannot be synchronously iterated and optimized; Fourth, the challenge of static system fixation, lack of independent upgrades, and compliance traceability: parameters are fixed after deployment of traditional solutions, and there is no full-process error summary and iteration; AI analysis is black-boxed and cannot meet the compliance requirements for power grid accident review.

[0024] Example 1: See Figure 1-3 This embodiment presents a DC control and protection auxiliary decision-making method based on multi-agent collaborative technology. It includes: constructing a hierarchical four-level business intelligent agent + global collaborative scheduling intelligent agent cluster, a three-level standardized Skill modular decoupling system, and a DC-specific adaptive coupling algorithm, which are deeply bound and linked to form an integrated closed-loop fault auxiliary decision-making system. Through mutual constraints and bidirectional parameter communication between the architecture layer, module layer, and algorithm layer, it collaboratively solves four major industry pain points: heterogeneous data adaptation at multiple DC sites, high-precision judgment of complex faults, full-link synchronous autonomous iteration, and full-process traceable and compliant operation and maintenance. It achieves collaborative technical gains that cannot be achieved by individual architectures, individual algorithms, or simple combinations of existing technologies.

[0025] The overall system utilizes a hierarchical, schedulable multi-agent cluster as the computing power carrier, a three-level standardized Skill module as the core of business decoupling, and five sets of coupled adaptive algorithms as the computational kernel. All algorithm execution parameters are distributed in real-time across agents via global collaborative scheduling. The entire process involves a five-stage closed-loop linkage, with each stage being inseparable. The specific implementation steps are as follows: Step 1: Fault Detection and Adaptive Processing of Multi-Source Data Standardization This step is completed by the data processing agent relying on a basic general-purpose Skill cluster. It adopts an agent-driven modular data adaptive preprocessing mechanism, coupled with a cross-site data adaptive normalization algorithm, to achieve unified standardized processing and dynamic adaptation of multi-source heterogeneous data. The algorithm running parameters issued to the data processing agent by the global collaborative scheduling agent include site adaptive correction coefficients.

[0026] The specific sub-steps are as follows: 1.1 Fault Classification Awareness: Through the event awareness skill, we continuously monitor alarm signals from the field business system, classify and identify various DC control and protection alarm information, prioritize the response to core fault alarms such as equipment abnormalities and protection actions, filter invalid disturbance signals, and ensure the pertinence and timeliness of fault response.

[0027] 1.2 Parallel Acquisition of Multi-Source Data: A multi-threaded parallel acquisition mechanism is established through the data acquisition skill to simultaneously capture DC control and protection fault waveform timing data and SER event text data, avoiding the problems of data timing misalignment and low acquisition efficiency in traditional serial acquisition methods.

[0028] 1.3 Adaptive Standardization Preprocessing: The data preprocessing skill is activated to sequentially complete data format unification, missing value completion, and redundant noise data cleaning. Addressing the issue of inconsistent data units and channel parameters across different converter stations, this invention employs a self-developed cross-site adaptive data normalization algorithm to achieve dynamic standardization transformation of the original data, adapting to the differentiated data characteristics of multiple stations.

[0029] The corresponding adaptive normalization calculation formula is as follows: , In the formula, The electrical quantity data is after standardization. The raw data collected on-site; These are the minimum and maximum historical operating data values ​​for the corresponding converter station s, respectively. The site adaptive correction coefficients are dynamically issued by the global collaborative scheduling agent and are output in real time by the upper-layer collaborative architecture, and cannot be statically preset.

[0030] 1.4 Time-Series Oriented Data Flow: Following a preset business interaction protocol, the data forwarding skill pushes the preprocessed standardized dataset to the downstream protection action analysis agent according to time-series logic. Simultaneously, all errors in this stage (data missing, normalization deviation) are transmitted back in real-time to the global collaborative scheduling agent for subsequent global iterations. calculate.

[0031] Step 2: Joint analysis and source tracing verification of protection actions based on bidirectional fusion of large and small models This step is the core innovation of this invention. The protection action analysis agent, relying on its specialized business skills, executes a self-developed bidirectional iterative fusion algorithm for both large and small models. This breaks away from the existing technical model of independent computation of large and small models and simple result superposition. Through a closed-loop mechanism of mutual optimization of dual model parameters and mutual calibration of results, it achieves dual technical improvements. The algorithm execution parameters issued to the protection action analysis agent by the global collaborative scheduling agent include threshold correction weights and dynamic fusion weights.

[0032] The specific sub-steps are as follows: 2.1 Scenario-based intelligent model matching: The model matching and calling Skill is based on fault semantic features and operating condition tags, and automatically matches a small power mechanism model that is suitable for the current fault scenario, replacing the traditional fixed model calling logic, and realizing accurate matching between fault scenarios and calculation models.

[0033] 2.2 Dual-model bidirectional iterative fusion operation: The original bidirectional iterative fusion mechanism of this invention realizes the parameter optimization of the small model by the large model and the deviation correction of the large model by the small model through two sets of self-developed formulas, forming a synergistic gain effect that a single model cannot achieve.

[0034] Firstly, the large model drives the adaptive optimization of parameters in the small model, which solves the problem that the fixed parameters of the traditional mechanism model cannot adapt to complex fault conditions (the first level of technical gain). The threshold optimization formula is as follows: , In the formula, The optimized protection action judgment threshold; The initial baseline threshold for the small mechanism model; The weights are dynamically adjusted by the threshold issued by the globally coordinated scheduling Skill based on the fault concurrency level; The value is the semantic feature score of the fault scene output by the large model, with a value range of [-1,1]. Only the multi-agent collaborative architecture can realize cross-module feature transfer.

[0035] Secondly, the small-scale mechanism model corrects the inference bias of the large-scale model, solving the problems of insufficient constraints on the power engineering mechanism and ambiguity in semantic reasoning in the large-scale model (a second layer of technical gain). The fusion correction formula is as follows: , In the formula, This is the final judgment result after the fusion of the two models; The semantic reasoning score for large models; The score is determined by the physical calculation of the small mechanism model; The weights are dynamically fused and adaptively allocated by the global scheduling unit based on the fault complexity. Based on the final judgment result after the fusion of the two models, a structured verification conclusion is generated, which includes the triggering sequence of protection actions, logical matching degree and deviation correction basis, to verify the compliance of protection actions.

[0036] 2.3 Full-process reasoning traceability: By breaking down the fault analysis logic layer by layer through multi-level thinking chain reasoning skills, the entire process of model parameter adjustment, rule comparison, and result correction data and judgment basis is fully recorded. This solves the black box problem of traditional AI analysis technology, builds a reproducible reasoning trajectory, and ensures that the analysis process is traceable and reproducible.

[0037] 2.4 Secondary verification of fusion results: The result verification Skill performs cross-verification on the judgment conclusions of the dual-model fusion output, removes abnormal deviation data generated by single model calculation, and sends the abnormal markers back to the global collaborative scheduling agent to ensure the accuracy and reliability of the protection action analysis results.

[0038] 2.5 Structured Result Flow and Archiving: Report Generation and Flow Skill automatically compiles and analyzes data, generates standardized protection action analysis reports, and synchronously pushes core characteristic parameters and fusion calculation results to the fault location intelligent agent, completing the closed-loop flow of upstream and downstream business data.

[0039] Step 3: Precise fault localization based on multi-feature coupling This step is executed by the fault location agent, relying on a self-developed multi-feature weighted fusion localization algorithm to integrate multi-dimensional fault assessment information, replacing the traditional single electrical feature judgment method, and achieving the dual technical effects of full coverage and high accuracy in fault location. The algorithm operation parameters issued to the fault location agent by the global collaborative scheduling agent include dynamic weight coefficients.

[0040] The specific sub-steps are as follows: 3.1 Initial Fault Scene Clustering: The initial fault judgment skill combines the upstream dual-model fusion analysis results to perform preliminary clustering judgment on fault types, narrowing the scope of fault judgment and providing basic support for accurate positioning.

[0041] 3.2 Differentiated Channel Data Matching: Based on the initial fault type, the channel data matching skill intelligently filters core analog and switching data such as three-phase voltage, current, differential current, and DC current, eliminating invalid data dimensions and improving the efficiency of feature analysis.

[0042] 3.3 In-depth analysis of electrical characteristics: The optimized mechanism model driven by the electrical quantity feature analysis skill extracts multi-dimensional core features such as electrical quantity amplitude, sudden change trend, fluctuation period, and phase deviation at the time of fault, constructs a standardized fault feature vector set, and completes the quantitative extraction of fault features.

[0043] 3.4 Multi-dimensional weighted coupling fault judgment: The present invention integrates four judgment dimensions: reasoning conclusions, multi-dimensional electrical characteristics, protection action logic, and regional topology characteristics. It has developed a multi-feature weighted fusion judgment formula to achieve accurate positioning through multi-information coupling and effectively avoid the limitations of single-dimensional judgment.

[0044] The multi-dimensional fault comprehensive judgment formula is as follows: , In the formula, The overall fault matching score is determined by the score; the higher the score, the higher the fault matching degree. The matching score is the result of dual-model fusion inference. The multidimensional electrical feature matching score; To protect the action logic matching score; The score is used to match regional topology faults. Dynamic weights are adaptively assigned to Skills for global collaborative scheduling, and the total weights are always 1.

[0045] 3.5 Multi-level reasoning and traceability: The multi-level thinking chain traceability skill fully records the reasoning steps and judgment basis of the entire process of fault initial judgment, feature extraction, multi-dimensional fusion analysis and fault location, so as to realize that the fault location results are verifiable and traceable.

[0046] 3.6 Standardized Result Output: Integrate the entire process judgment data and reasoning logs, unify the data format, and push it to the fault handling plan generation intelligent agent to support the intelligent compilation of subsequent operation and maintenance plans.

[0047] Step 4: Intelligent solution generation through multi-dimensional case matching and operation and maintenance data fusion This step is executed by the intelligent agent that generates the fault handling plan. Through a self-developed multi-dimensional case similarity matching algorithm, it achieves cross-domain fusion of fault technical characteristics and operation and maintenance business data, improving the accuracy of case matching and the practicality of the repair plan. The specific sub-steps are as follows: 4.1 Multi-dimensional Similar Case Intelligent Retrieval: Using fault type, fault location, electrical characteristics, and model inference parameters as composite search tags, this method abandons the traditional single-tag retrieval approach and achieves accurate matching of historical cases based on the Euclidean distance similarity algorithm. The algorithm operation parameters issued by the global collaborative scheduling agent to the fault location agent include fault feature weight coefficients.

[0048] The formula for calculating case similarity matching is as follows: , In the formula, This represents the similarity between the current fault and historical cases; a smaller value indicates higher matching accuracy. These are the weighting coefficients for fault characteristics in each dimension; The i-th dimension feature parameter of the current fault; The i-th dimension feature parameter of historical failure cases enables accurate matching and comparison of multi-dimensional failure features.

[0049] 4.2 Intelligent Orchestration of Operation and Maintenance Data: By combining multi-dimensional business data such as fault severity level, historical best maintenance plan, on-duty schedule of operation and maintenance personnel, and operating conditions of site equipment, an inspection plan and maintenance scheme adapted to the current fault are intelligently generated, realizing the deep integration of technical analysis and on-site operation and maintenance business.

[0050] 4.3 Targeted push of handling information: Fault analysis reports, similar fault cases, and customized maintenance plans are pushed to the on-site operation and maintenance terminal simultaneously to provide data support for front-line fault handling.

[0051] 4.4 Full-process data iteration and archiving: The original data, model parameters, inference logs, and repair plans of this failure are fully archived to the business database, and the historical case library and model training samples are updated synchronously to provide data support for the system's autonomous iteration and optimization.

[0052] Step 5: Cross-scenario adaptive and autonomous iterative optimization of global collaborative scheduling This step involves the global collaborative scheduling Skill coordinating the entire system's operation. Through optimization of the agent's global scheduling architecture and a dynamic iteration algorithm for Skill parameters, the system achieves adaptive adaptation and autonomous upgrades, overcoming the technical shortcomings of traditional systems' fixed operation and lack of iterative capabilities. The iterative formula loss value... To summarize the full-process errors of four types of business intelligent agents, which cannot be obtained independently by a single intelligent agent or a single algorithm, the specific sub-steps are as follows: 5.1 Process Sequence Control: The process sequence control Skill standardizes the triggering order and data interaction logic of each agent and Skill unit, avoiding process disorder and data conflict caused by the parallel operation of multiple modules, and ensuring the stable and orderly operation of the entire process.

[0053] 5.2 Cross-site dynamic adaptive adaptation: The dynamic adaptation skill reads the configuration files of each converter station and dynamically adjusts the data preprocessing rules, model fusion weights, and fault judgment thresholds based on the self-developed skill parameter iteration formula, without the need for manual intervention and station-by-station customized development.

[0054] 5.3 Fault Tolerance Handling: The Fault Tolerance Skill monitors the running status of all system modules in real time. For faults such as data interruption, model call failure, and inference anomalies, it automatically triggers retry mechanisms and anomaly reporting processes to ensure uninterrupted and stable system operation.

[0055] 5.4 Full-dimensional and global autonomous iterative upgrade: Based on long-term archived full-volume fault data, continuously optimize Skill scheduling logic, large and small model fusion weights, multi-feature judgment rules and case matching algorithms to achieve autonomous iterative upgrade of the system's core technical capabilities, which is different from the existing static and fixed technical solutions.

[0056] , In the formula, This includes all Skills, models, and judgment weight parameters after iterative updates; These are the baseline parameters for the k-th fault operation; The iterative learning rate is used to control the magnitude of parameter updates. The weighted sum of error loss values ​​for the four stages of data processing, protection analysis, fault location, and case matching can only be aggregated and calculated by the globally coordinated scheduling agent. After the iteration is completed, the algorithm running parameters, the built-in skills of all agents, and the model operation parameters are updated and distributed to achieve synchronous and autonomous upgrade of the entire system.

[0057] Example 2: See Figure 1-3 This embodiment provides a DC control and protection auxiliary decision-making system based on multi-agent cooperative technology, applied to the method described in Embodiment 1, which includes: The system includes a data processing agent, a protection action analysis agent, a fault location agent, a fault handling plan generation agent, and a global collaborative scheduling agent, and is equipped with three levels of standardized skill units: basic general skills, specialized business skills, and collaborative scheduling skills. The global collaborative scheduling agent is configured to coordinate the operation of the entire system and dynamically distribute algorithm running parameters to each business agent and Skill unit; The data processing agent is used to execute basic general skills, perform adaptive standardization preprocessing on the collected multi-source heterogeneous fault data based on the corresponding algorithm running parameters, and push the processed standardized data to the protection action analysis agent according to the time sequence logic. The protection action analysis agent is used to execute a special business skill, call the bidirectional iterative fusion algorithm of large and small models, perform joint analysis and source verification of protection actions based on the corresponding algorithm running parameters, and output the dual-model fusion judgment result containing model inference parameters. The fault location intelligent agent is used to execute a special business Skill, call a multi-feature weighted fusion location algorithm, and perform weighted summation calculation on the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score based on the corresponding algorithm running parameters to obtain a comprehensive fault matching score. The fault handling solution generates an intelligent agent to execute a special business skill. It constructs multi-dimensional search tags based on the model inference parameters, fault type, fault location, and electrical characteristics. It determines the case matching weight based on the corresponding algorithm running parameters, calls the multi-dimensional similar case matching algorithm to retrieve historical cases, extracts standard maintenance steps from similar cases, and generates a maintenance plan by combining the fault severity level and operation and maintenance resource configuration data. The global collaborative scheduling agent is used to summarize the weighted sum of error loss values ​​of the four stages of data processing, protection action analysis, fault location and case matching. Based on the Skill parameter dynamic iteration algorithm, it synchronously and autonomously updates the parameters of the entire system. The algorithm running parameters are part of the parameters of the entire system, and the corresponding algorithm running parameters are distributed to each business agent and Skill unit.

[0058] The overall technical solution of this invention achieves substantial innovation through architecture reconstruction, self-developed algorithms, and deep coupling of multiple technologies. Each modular skill unit can be flexibly combined and called according to the fault scenario. The core fusion algorithm, iterative algorithm, and positioning algorithm are all designed specifically for DC control and protection fault scenarios. There is no simple transplantation and adaptation of existing technologies. Multiple superimposed technical effects are achieved through the linkage of multiple technologies, which effectively solves various defects of existing technologies. At the same time, it meets the operation and maintenance compliance requirements of the power industry by relying on the full-process traceability mechanism.

[0059] Technical effects of this embodiment: 1. Compared to existing technologies, this invention abandons conventional process optimization and local parameter adaptation methods. Through original architecture design and self-developed algorithm system, it achieves multi-dimensional technological breakthroughs and synergistic gains. Specific beneficial effects are as follows: First, it enables rapid deployment across converter stations without customization, significantly reducing project implementation cycle and lowering implementation costs. This invention constructs a global collaborative scheduling system, equipped with a three-level decoupled Skill functional module, which can realize the core algorithm parameter α. s The system dynamically distributes and adaptively corrects and optimizes β, ω, and ω1~ω4. For new access scenarios at UHV converter stations, only the corresponding equipment configuration file needs to be uploaded, and the system can autonomously adapt to the entire fault analysis process without requiring customized development and on-site debugging for individual stations. Compared to the traditional 30-day deployment cycle for a single station, this technology can reduce the project deployment time to 5-7 days, with an overall implementation cost reduction of over 80%. This superior performance is not achieved solely through a multi-agent architecture or normalization algorithm, but rather through the unique technological gains resulting from the deep coupling and synergistic empowerment of multiple core technologies.

[0060] 2. Breaking through the technical bottlenecks of complex DC fault assessment, comprehensively improving fault location and assessment accuracy. This invention employs a bidirectional mutually optimized coupling algorithm with large and small models, effectively compensating for the shortcomings of single models in complex fault scenarios. Simultaneously, it introduces a multi-feature dynamic weighted fusion positioning algorithm to integrate multi-dimensional core feature information of DC faults, constructing a comprehensive and high-precision fault assessment system. This is particularly effective for complex operating conditions such as valve short circuits, commutation failures, high-resistance grounding, and multiple concurrent faults. Such high-precision assessment results cannot be achieved by relying solely on single algorithm optimization or traditional architecture iteration, fully highlighting the core advantages of multi-technology integration and innovation.

[0061] 3. Establish a system-wide, globally synchronized, and autonomous iterative mechanism to ensure stable and consistent system performance over the long term. This invention innovates a globally iterative computation logic, uniformly aggregating the operational errors ΔE of the four major business intelligent agents throughout the entire process through iterative formulas. This allows for a one-time, simultaneous iterative update of data preprocessing rules, model judgment thresholds, fault assessment weights, and case matching coefficients. Engineering testing has verified that after 10 months of continuous and stable operation, the fault assessment accuracy showed no significant decline. In contrast, existing traditional technologies can only achieve fragmented iterative optimization of single models and local parameters, which is prone to continuous performance degradation over long-term operation. The effects of this globally synchronized iteration and long-term stable performance cannot be achieved through existing single technologies or simple combinations of conventional technologies.

[0062] 4. Constructing a full-link traceability technology system to fully meet the stringent requirements of compliant operation and maintenance in the power industry. This invention embeds a multi-level thought chain traceability Skill module into all intelligent agents and functional units, which can completely retain core data and judgment criteria throughout the entire process, including cross-agent parameter interaction, formula calculation and derivation, and dynamic weight allocation. It completely solves the industry pain points of black boxing of general AI technology, unverifiable operation process, and lack of retention of judgment criteria, and is fully compatible with the standards and norms of power grid accident review and safety compliance verification. Currently, the three mainstream technical architectures in the industry—large model + general multi-agent collaborative scheduling, fusion of large and small models and local self-iteration, and multi-agent + knowledge graph + large model power fault diagnosis—do not have a dedicated full-link traceability mechanism adapted to DC fault scenarios. This technology has achieved a key technological breakthrough.

[0063] Example 3: An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the method as described in any one of Examples 1.

[0064] Example 4: A computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the method as described in any one of Examples 1.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A DC control and protection auxiliary decision-making method based on multi-agent cooperative technology, characterized in that, include: S1. The data processing agent executes the basic general skill, performs adaptive standardization preprocessing on the collected multi-source heterogeneous fault data based on the corresponding algorithm running parameters, and pushes the processed standardized data to the protection action analysis agent according to the time sequence logic. S2, the protection action analysis agent executes a special business skill, a bidirectional iterative fusion algorithm of large and small models, performs joint analysis and source verification of protection actions based on the corresponding algorithm running parameters, and outputs a dual-model fusion judgment result containing model inference parameters; S3. The fault location agent executes a special business Skill, calls a multi-feature weighted fusion location algorithm, and performs a weighted summation calculation on the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score based on the corresponding algorithm running parameters to obtain a comprehensive fault matching score. S4. The fault handling solution generation intelligent agent executes the special business Skill, constructs multi-dimensional search tags based on the model inference parameters, fault type, fault location, and electrical characteristics, determines the case matching weight based on the corresponding algorithm running parameters, calls the multi-dimensional similar case matching algorithm to retrieve historical cases, extracts the standard maintenance steps in similar cases, and generates a maintenance plan by combining the fault severity level and operation and maintenance resource configuration data. S5, the global collaborative scheduling intelligent agent summarizes the weighted sum of the error loss values ​​of the four stages of data processing, protection action analysis, fault location and case matching, updates the parameters of the whole system based on the Skill parameter dynamic iterative algorithm, and sends the corresponding algorithm running parameters to each business intelligent agent and Skill unit.

2. The DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in claim 1, characterized in that, The basic general skills include event awareness skills, data acquisition skills, data preprocessing skills, and data forwarding skills; Step S1 includes: S1.1 The data processing intelligent agent continuously monitors the alarm signals of the field business system through the event perception Skill, classifies and identifies DC control and protection alarm information, and prioritizes responding to core alarms of equipment abnormality and protection action. S1.2 Establish a multi-threaded parallel acquisition mechanism through the data acquisition skill to synchronously capture DC control and protection fault waveform timing data and SER event text data; S1.

3. Start the data preprocessing skill and use the cross-site data adaptive normalization algorithm to standardize the collected raw electrical quantity data. The operation of the cross-site data adaptive normalization algorithm is based on the algorithm operation parameters issued by the global collaborative scheduling agent. S1.4 Data forwarding Skill follows the preset business interaction protocol, pushes the pre-processed standardized dataset to the downstream protection action analysis agent according to the time sequence logic, and transmits the full process error of this stage back to the global collaborative scheduling agent in real time for global iterative calculation.

3. The DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in claim 1, characterized in that, Step S2 includes: S2.1 The protection action analysis intelligent agent automatically matches a small power mechanism model that is compatible with the current fault scenario based on fault semantic features and operating condition labels through a special business skill. S2.2 Perform bidirectional iterative fusion calculation of the size model, the calculation including: S2.

21. Using the output of the large model and the threshold correction weight issued by the global collaborative scheduling agent, the initial baseline threshold of the mechanism small model is dynamically optimized to obtain the optimized protection action judgment threshold. S2.

22. Using the physical calculation results of the mechanism small model and the dynamic fusion weights issued by the global collaborative scheduling agent, the semantic reasoning judgment score of the large model is corrected for deviation; the final judgment result after the fusion of the two models is output. S2.

23. Based on the final judgment result after the fusion of the two models, generate a structured verification conclusion that includes the protection action triggering sequence, logic matching degree and deviation correction basis to verify the compliance of the protection action; S2.

3. Through multi-level thinking chain reasoning skills, the fault analysis logic is broken down layer by layer, and the data and judgment basis of the whole process of model parameter adjustment, rule comparison and result correction are fully recorded to construct a reproducible reasoning trajectory. S2.

4. The judgment conclusion of the dual-model fusion output is cross-validated by the result verification Skill, abnormal deviation data generated by single model operation is removed, and the abnormal marker is sent back to the global collaborative scheduling agent. S2.

5. Compile and analyze the data, generate a standardized protection action analysis report, and push the core fault characteristics and fusion judgment results to the downstream fault location intelligent agent according to the time sequence logic. At the same time, the calculation error of this link is sent back to the global collaborative scheduling intelligent agent for global iterative calculation.

4. The DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in claim 1, characterized in that, Step S3 includes: S3.1 The fault location intelligence receives the dual-model fusion analysis conclusion issued by the upstream protection action analysis intelligence. The fault preliminary judgment Skill combines the upstream dual-model fusion analysis conclusion to perform preliminary judgment clustering of fault types and narrow the scope of fault judgment. S3.2 Channel Data Matching Skill: Based on the initial fault type, intelligently filter the core data of three-phase voltage, current, differential current and switching quantity associated with the fault, and eliminate invalid data dimensions. S3.3 Electrical quantity characteristic analysis: The mechanism mini-model driven by Skill extracts the amplitude, sudden change trend, fluctuation period and phase deviation of electrical quantities at the time of fault, and constructs a standardized fault feature vector set. S3.

4. Call the multi-feature weighted fusion localization algorithm. Based on the dynamic weight coefficients issued by the global collaborative scheduling agent, perform weighted summation calculation on the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score to obtain the comprehensive fault matching score and lock the specific location and type of the fault. S3.

5. Through multi-level thinking chain tracing skill, the entire process of reasoning steps and calculation parameters of fault initial judgment, feature extraction, multi-dimensional fusion analysis and result determination are fully recorded to support the verification and tracing of fault location results. S3.6 Integrate the entire process analysis data and inference logs, unify the data format, and push it to the fault handling solution generation agent. The positioning error of this step is transmitted back to the global collaborative scheduling agent to participate in the global parameter iterative optimization.

5. The DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in claim 1, characterized in that, The fault handling scheme generating agent receives standardized localization results and inference logs from the fault location agent. The standardized localization results include fault type, fault location, and model inference parameters generated by the protection action analysis agent. Step S4 includes: S4.1 Extract multi-dimensional retrieval tags based on the fault type, fault location, electrical characteristics, and model inference parameters generated by the protection action analysis agent, and construct a retrieval vector; call the case similarity matching algorithm, and calculate the similarity between the retrieval vector and the feature vector in the historical fault case library based on the weight coefficients dynamically issued by the global collaborative scheduling agent; S4.2 Based on the similar cases obtained from the search, extract the standard maintenance steps in the case; combine the current fault severity level, the operating status of equipment in the station and the operation and maintenance resource configuration data to generate an inspection plan and maintenance scheme adapted to this fault. S4.3 Push the generated fault analysis report, similar fault cases and customized maintenance plan to the field operation and maintenance terminal to provide data support for front-line fault handling; S4.4 Archive all original data, model parameters, inference logs, and generated repair plans for this fault to the business database, and send the calculation errors in the case matching process back to the global collaborative scheduling agent, and update the historical case library and model training sample set simultaneously to provide data support for the autonomous iterative optimization of the global collaborative scheduling agent.

6. The DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in claim 1, characterized in that, Step S5 includes: S5.

1. By controlling the timing of processes through Skills, the triggering order and data interaction protocol of each intelligent agent and Skill unit are standardized, and the timing constraints for cross-module collaborative execution are established. S5.2 reads the configuration files of each converter station by dynamically adapting the Skill and dynamically iterates the algorithm based on the Skill parameters to adjust the data preprocessing rules, model fusion weights and fault judgment thresholds in order to adapt to the differentiated operating environment of different converter stations. S5.

3. Real-time monitoring of the operating status of all system modules through the fault tolerance skill. Automatically trigger retry mechanism, parameter rollback mechanism or abnormal reporting process for data interruption, model call failure or inference abnormal conditions. S5.4 summarizes the full-process error loss value of the four stages of data processing, protection action analysis, fault location and case matching, assigns corresponding weighting coefficients to each component and sums them to obtain the global loss value; based on the Skill parameter dynamic iteration algorithm, the global loss value is used to synchronously update and distribute the algorithm running parameters, all Skill scheduling logic, model operation parameters and judgment weights in the whole system.

7. A DC control and protection auxiliary decision-making system based on multi-agent cooperative technology, applied to the DC control and protection auxiliary decision-making method based on multi-agent cooperative technology as described in any one of claims 1-6, characterized in that, include: The system includes a data processing agent, a protection action analysis agent, a fault location agent, a fault handling plan generation agent, and a global collaborative scheduling agent, and is equipped with three levels of standardized skill units: basic general skills, specialized business skills, and collaborative scheduling skills. The global collaborative scheduling agent is configured to coordinate the operation of the entire system and dynamically distribute algorithm running parameters to each business agent and Skill unit; The data processing agent is used to execute a basic general skill, perform adaptive standardization preprocessing on the collected multi-source heterogeneous fault data based on the corresponding algorithm running parameters, and push the processed standardized data to the protection action analysis agent according to the time sequence logic. The protection action analysis agent is used to execute a special business skill, call the bidirectional iterative fusion algorithm of large and small models, perform joint analysis and source verification of protection actions based on the corresponding algorithm running parameters, and output the dual-model fusion judgment result containing model inference parameters. The fault location intelligent agent is used to execute a special business Skill, call a multi-feature weighted fusion location algorithm, and perform weighted summation calculation on the dual-model fusion inference score, multi-dimensional electrical feature score, protection action logic score and regional topology fault matching score based on the corresponding algorithm running parameters to obtain a comprehensive fault matching score. The fault handling solution generates an intelligent agent to execute a special business skill. It constructs multi-dimensional search tags based on the model inference parameters, fault type, fault location, and electrical characteristics. It determines the case matching weight based on the corresponding algorithm running parameters, calls the multi-dimensional similar case matching algorithm to retrieve historical cases, extracts standard maintenance steps from similar cases, and generates a maintenance plan by combining the fault severity level and operation and maintenance resource configuration data. The global collaborative scheduling agent is used to summarize the weighted sum of error loss values ​​across the four stages of data processing, protection action analysis, fault location, and case matching. Based on the Skill parameter dynamic iteration algorithm, it synchronously and autonomously updates the parameters of the entire system and distributes the corresponding algorithm running parameters to each business agent and Skill unit.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images