Charging pile operation and maintenance method and system based on multi-agent cooperation

Abstract

The invention discloses a charging pile operation and maintenance method and system based on multi-agent cooperation. The method comprises the following steps: configuring an MCP server corresponding to each functional subsystem, and packaging heterogeneous rear ends into standard resources and tools; configuring an expert agent to call resources through an MCP protocol so as to carry out anomaly perception; when external cooperation is needed, the agents negotiate through an A2A protocol or send a request to an arbitration agent; and the arbitration agent organizes multiple rounds of structured message exchange to jointly diagnose and judge the fault root cause, thereby generating an operation and maintenance work order. According to the method, the problem that heterogeneous equipment is difficult to access is solved by utilizing the MCP, the subsystem barrier is broken through the A2A protocol and an arbitration mechanism, and accurate collaborative diagnosis of cross-domain complex faults is realized.

Description

Technical Field

[0001] This invention relates to the field of charging pile operation and maintenance technology, and in particular to a charging pile operation and maintenance method and system based on multi-agent collaboration. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the scale of charging infrastructure construction is expanding daily. As an industrial device integrating multiple complex functional subsystems such as power electronics, thermal management, electrical connection, and embedded control, the operational stability of charging piles directly affects the user's charging experience and power grid safety.

[0003] In the existing charging pile operation and maintenance model, various functional subsystems (such as charging modules, BMS, and cooling systems) are typically provided by different suppliers. Due to reasons such as trade secrets, suppliers often do not fully disclose their internal design drawings, underlying documents, or detailed fault mode libraries. To achieve fault diagnosis, each supplier may train its own dedicated intelligent agents or algorithm models to monitor the components they provide. These intelligent agents are usually deployed on cloud platforms or deployed locally on the charging pile in a lightweight form.

[0004] However, in the actual process of intelligent operation and maintenance of charging piles, the current system architecture generally suffers from the following two major technical bottlenecks.

[0005] First, there is a knowledge access gap. The data and knowledge required for charging pile operation and maintenance are scattered across various heterogeneous systems, such as Supervisory Control and Data Acquisition (SCADA) systems, Battery Management Systems (BMS), Enterprise Resource Planning (ERP) systems, and various unstructured knowledge bases and document servers. Existing technologies struggle to dynamically and systematically access this multimodal operation and maintenance knowledge (including real-time data, technical documents, API interfaces, etc.), preventing the operation and maintenance system from fully utilizing existing data assets for in-depth analysis.

[0006] Second, there is a gap in agent collaboration. Because different component manufacturers or agents targeting different functions (e.g., charging module agents, thermal management agents) have relatively independent and closed knowledge domains, they lack a unified communication language and negotiation mechanism. When encountering complex faults across subsystems (e.g., power module derating due to poor heat dissipation), each agent often can only make isolated judgments based on its own data, making it difficult to conduct structured communication, evidence exchange, and joint consultations like a human expert team. This leads to rigid collaboration, hindering truly dynamic, multi-objective collaborative decision-making and making it difficult to identify the root cause of the fault.

[0007] In response, existing integration solutions often employ custom APIs or centralized databases for hard-coded integration. This approach not only results in high system coupling and poor scalability, but also fails to achieve ideal practical results when faced with massive amounts of heterogeneous data and complex collaborative logic. Summary of the Invention

[0008] The purpose of this invention is to provide a charging pile operation and maintenance method and system based on multi-agent collaboration, which can achieve flexible and secure interaction between intelligent agents and massive knowledge sources, and standardize the complex collaboration process between intelligent agents, in order to solve the above-mentioned technical problems.

[0009] To achieve the above objectives, this invention provides a charging pile operation and maintenance method based on multi-agent collaboration, comprising: Configure multiple Application Model Context Protocol (MCP) servers corresponding to each functional subsystem in the charging pile. The MCP servers are used to encapsulate the data interfaces and functions of the back-end heterogeneous systems related to each functional subsystem into resources and tools that conform to the MCP protocol standard. Multiple expert agents are configured, each corresponding to a specific MCP server. The expert agents are configured to invoke the resources and tools provided by the MCP server through the MCP protocol to obtain standardized data and perform functional subsystem-level anomaly detection and preliminary diagnosis. When any of the aforementioned expert agents perceives an anomaly and determines that external collaboration is required, it conducts bilateral negotiations with other expert agents associated with the anomaly through the inter-agent communication protocol A2A, and sends an arbitration request message to the arbitration agent when it is determined that multi-party consultation is required. After receiving the arbitration request message, the arbitration agent organizes all relevant expert agents through the A2A protocol to conduct multiple rounds of structured message exchange in order to jointly diagnose and determine the root cause of the fault. A maintenance work order is generated based on the root cause of the fault.

[0010] Preferably, the expert agent subscribes to the data streams it is interested in through the MCP server for routine monitoring, and calls the tools provided by the MCP server to analyze the data streams in order to detect anomalies and generate functional subsystem-level abnormal events.

[0011] Preferably, the two expert agents achieve bilateral negotiation by sending structured A2A messages, which include at least one of intent, context, content, and constraints.

[0012] Preferably, after receiving the bilateral negotiation message, other expert agents associated with the anomaly invoke tools and query resources through their corresponding MCP servers to collect evidence, and reply with analysis results through the A2A protocol. The analysis results include at least one of intent, content, and suggestions.

[0013] Preferably, the arbitration agent organizes all relevant expert agents to conduct multi-round structured message exchanges, including sending one or more of the following: invitation to join message, request for evidence message, question message, receiving evidence provision message, submit analysis message, and submit simulation message.

[0014] Preferably, the arbitration agent sends the arbitration result to the process orchestration agent via the A2A protocol; after receiving the arbitration result, the process orchestration agent generates the maintenance work order based on the root cause of the fault.

[0015] Preferably, the process orchestration agent sends the maintenance work order to the default execution agent based on the complexity of the maintenance work order, or initiates a task bidding process based on the A2A protocol to send the maintenance work order to the winning execution agent.

[0016] Preferably, the winning execution agent executes the tasks in the maintenance work order and synchronizes the task status with all relevant expert agents in real time through the A2A protocol; the relevant expert agents verify the post-maintenance data through the MCP protocol and provide feedback on the verification results through the A2A protocol.

[0017] Preferably, after the maintenance work order is closed, the A2A message stream and MCP call record generated during this maintenance are stored in the knowledge base; the prediction model of the relevant expert agent is incrementally trained and optimized through the tools managed by the MCP server.

[0018] The present invention also provides a charging pile operation and maintenance system based on multi-agent collaboration. The operation and maintenance system works based on the charging pile operation and maintenance method described above to perform fault maintenance on the charging pile.

[0019] The present invention also provides a charging pile operation and maintenance system, which includes: One or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the charging pile operation and maintenance method as described above.

[0020] The present invention also provides a computer-readable storage medium, characterized in that it includes a computer program, which can be executed by a processor to perform the charging pile operation and maintenance method as described above.

[0021] Compared with existing technologies, the charging pile operation and maintenance method provided by the above-mentioned technical solution of the present invention effectively solves the knowledge access gap by configuring MCP server to encapsulate heterogeneous back-end systems as standard resources, and realizes standardized dynamic access and efficient utilization of multimodal operation and maintenance knowledge. At the same time, by using A2A protocol and arbitration mechanism to organize expert intelligent agents to conduct structured negotiation and joint diagnosis, it effectively solves the intelligent agent collaboration gap, breaks down the isolation barriers between various functional subsystems in the charging pile, and realizes accurate collaborative diagnosis of cross-domain complex faults like a human expert team. Attached Figure Description

[0022] Figure 1 This is a diagram of the charging pile operation and maintenance system architecture in an embodiment of the present invention.

[0023] Figure 2 This is a flowchart of a charging pile operation and maintenance method in one embodiment of the present invention.

[0024] Figure 3 This is a diagram illustrating the diagnostic negotiation sequence of a charging pile operation and maintenance method in one embodiment of the present invention.

[0025] Figure 4 This is a flowchart of the task bidding process in one embodiment of the present invention.

[0026] Figure 5 This is a flowchart illustrating the overall process of the charging pile operation and maintenance method of the present invention. Detailed Implementation

[0027] To illustrate the technical content, structural features, objectives, and effects of the present invention in detail, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0028] This embodiment discloses a charging pile operation and maintenance method based on multi-agent collaboration. The system architecture corresponding to this method is constructed based on a dual-protocol stack of the Application Model Context Protocol (MCP) and the Agent-to-Agent Communication Protocol (A2A). Figure 1 As shown, the overall system architecture is divided into a perception layer, an MCP protocol layer, an A2A protocol layer, and an execution and feedback layer from bottom to top. This architecture aims to solve the problems of difficulty in connecting heterogeneous devices and barriers to intelligent agent collaboration.

[0029] Based on this, please refer to the following: Figure 1 and Figure 2 The operation and maintenance method for this charging pile includes the following steps: S100: Configures the MCP server to encapsulate heterogeneous system data and functions.

[0030] Multiple MCP servers are configured at the MCP protocol layer, each corresponding to a different functional subsystem in the charging pile (such as the charging module subsystem, thermal management subsystem, electrical connection subsystem, etc.). For example, there are charging module MCP servers, electrical connection MCP servers, thermal management MCP servers, and control board MCP servers.

[0031] The MCP server acts as middleware, connecting downwards to the various functional subsystems and their related heterogeneous backend systems. In this embodiment, the MCP server is configured to encapsulate and map the raw data interfaces (such as Modbus register addresses, SQL query interfaces, and private APIs) and functional instructions of the backend heterogeneous systems into resources and tools that conform to the MCP protocol standard.

[0032] Resources: refers to passive, readable data entities, such as real-time voltage and current waveforms, historical log files, and equipment technical manuals (PDFs).

[0033] Tools: Refers to executable functions or scripts, such as executing a harmonic analysis algorithm, querying temperature curves for a specific time period, or issuing reset commands.

[0034] S200: Configures an expert intelligent agent to detect anomalies through the MCP server.

[0035] Multiple expert agents are configured at the A2A protocol layer, each corresponding to a different MCP server. These include, for example, a charging module agent, an electrical connection agent, a thermal management agent, and a control board agent. The expert agents connect to their respective MCP servers via the MCP protocol client interface.

[0036] Expert agents are configured to periodically or event-drivenly invoke resources and tools provided by the MCP server. Through the MCP server, expert agents acquire standardized data and process it using encapsulated analysis tools, thereby enabling them to detect and preliminarily diagnose anomalies in their respective responsible functional subsystems. For example, the charging module agent detects abnormal output power fluctuations using tools encapsulated on the MCP server.

[0037] S300: Bilateral negotiation and arbitration requests based on A2A agreements.

[0038] When any expert agent (e.g., the charging module agent) senses an anomaly and determines that external cooperation is needed, it conducts bilateral negotiations with other expert agents (e.g., the thermal management agent) that are associated with the anomaly through the A2A protocol.

[0039] If no agreement can be reached during bilateral negotiations or if the fault is found to involve more subsystems, and it is determined that a multi-party consultation is required, the initiating party or collaborating party shall send an arbitration request message to the arbitration agent through the A2A protocol.

[0040] S400: Arbitration agent organization for joint diagnosis and adjudication.

[0041] After receiving the arbitration request message, the arbitration agent organizes all relevant expert agents to conduct multiple rounds of structured message exchange through the A2A protocol. Specifically, this includes sending one or more of the following messages: inviting third parties (such as electrical connection agents) to join, requesting evidence, questioning, receiving evidence provision, submitting analysis, and submitting simulation, in order to conduct joint diagnosis and ultimately determine the root cause of the fault.

[0042] S500: Generates maintenance work orders based on the root cause of the fault, including fault location, recommended repair solutions, and required spare parts, so that they can proceed to the subsequent execution process.

[0043] Based on the charging pile operation and maintenance method in this embodiment, by configuring the MCP server to encapsulate heterogeneous backend systems as standard resources, the knowledge access gap is effectively solved, and the standardized dynamic access and efficient utilization of multimodal operation and maintenance knowledge are realized. At the same time, by using the A2A protocol and arbitration mechanism to organize expert agents to conduct structured negotiation and joint diagnosis, the collaboration gap of the agents is effectively solved, the isolation barriers between various functional subsystems in the charging pile are broken, and the accurate collaborative diagnosis of cross-domain complex faults is realized, just like a human expert team.

[0044] In another embodiment, the MCP server acts as a bridge connecting the perception layer and the A2A protocol layer, offering flexible deployment options. Specifically, the MCP server can be deployed on a cloud platform, edge computing node, or as a local embedded system of a charging pile.

[0045] For the resource and tool encapsulation of the MCP server, taking the charging module MCP server as an example, it connects downwards to heterogeneous backend systems such as SCADA real-time library, file server, fault database, simulation platform, and prediction model service, and provides standard interfaces upwards.

[0046] The resource encapsulation configuration is as follows: The current and voltage data acquired by SCADA are encapsulated into a streaming data stream, such as: resource: / / waveform / realtime; Package the historical efficiency curves into a CSV file; Package the IGBT technical manual in the file server into a PDF document, such as resource: / / docs / igbt_manual.pdf; The failure case library is encapsulated as a knowledge graph resource.

[0047] The tool encapsulation configuration is as follows: Package the harmonic analysis tool that runs locally into a Python script; The cloud-based junction temperature prediction model is encapsulated as an ML API.

[0048] The capacitor ESR (Equivalent Series Resistance) calculator is packaged as an algorithm service.

[0049] The parameter tuning simulator is packaged into a simulation environment.

[0050] In addition, the expert agent (i.e., the MCP client) subscribes to the data streams it is interested in (such as real-time current waveforms) through the MCP server for routine monitoring. The expert agent automatically calls the tools provided by the MCP server (such as harmonic analysis tools) to analyze the data streams, detect anomalies, and generate functional subsystem-level abnormal events.

[0051] For example, if the total harmonic distortion (THD) value returned by the harmonic analysis tool exceeds a preset threshold (e.g., 5%), or if the predicted temperature rise shown by the junction temperature prediction model exceeds the safety margin (e.g., ΔT>10°C), the expert agent generates a functional subsystem-level abnormal event, triggering subsequent processes.

[0052] The following specific example illustrates the anomaly negotiation and diagnosis process in the charging pile operation and maintenance method of this invention. For example... Figure 3 This example details the complete process of how expert agents can use the A2A protocol to conduct bilateral negotiation, escalation arbitration, multi-party debate, and final ruling when a charging station experiences a complex fault (such as unstable output power accompanied by abnormal heat generation).

[0053] Assuming the charging station is operational, the charging module intelligent agent, thermal management intelligent agent, electrical connection intelligent agent, and arbitration intelligent agent have all been activated. Each expert intelligent agent monitors the data flow of its respective subsystem in real time through the MCP server. The specific execution flow is as follows.

[0054] S10: Anomaly detection and preliminary investigation.

[0055] The charging module's intelligent agent calls the harmonic analysis tool through the MCP server, detects an increase in the output current harmonics of a certain charging module, predicts an increase in junction temperature, and generates an abnormal event.

[0056] The charging module agent determines that the anomaly may be related to heat dissipation, and therefore, as the initiator, sends structured message A to the thermal management agent via the A2A protocol. The specific content of message A is as follows: Intent: To assist in diagnosis; Context: The third harmonic of module 3 increases, predicting an increase in junction temperature; Content: Please provide the current air intake velocity and radiator temperature difference; Constraint: Response time < 2000ms.

[0057] After receiving message A, the thermal management agent queries local sensor data through its MCP interface and replies with message B within the constraint time. The content of message B is as follows: Intent: Information feedback; Contents: Wind speed 60% (rated value), temperature difference ΔT = 8°C (exceeds the preset threshold of 7°C); Recommendation: Suspected insufficient heat dissipation performance.

[0058] S11: Escalation of Disputes and Arbitration Intervention.

[0059] Based on its own data and feedback from the thermal management agent, the charging module's intelligent agent concluded that although the temperature difference was large, it could not rule out the possibility that loose electrical connections were causing the overheating and thus triggering harmonics. It also could not determine whether the problem was a fan malfunction or a blocked airflow. Given that the fault spanned multiple subsystems and was interconnected, the charging module's intelligent agent decided to initiate arbitration.

[0060] Then, the charging module agent sends message C to the arbitration agent. The content of message C is as follows: Intent: To initiate a debate; Context: Charging module malfunction combined with heat dissipation malfunction; Content: Request the organization to conduct cross-component diagnostics to determine the root cause of the fault.

[0061] The thermal management agent detects that an arbitration session has started and proactively sends message D to join the session. The content of message D is as follows: Intent: To participate in the diagnosis; Content: Supporting the view that there is insufficient heat dissipation, requesting the submission of evidence.

[0062] S12: Introduction of third-party evidence and diagnosis by exclusion method.

[0063] The arbitral agent takes over the session. To eliminate the potential interference of loose electrical connections, it sends message E to the electrical agent. The content of message E is as follows: Intent: To request evidence; Context: Troubleshooting module overheating issues; Content: Please check the status of the relevant electrical connection points; Constraint: Contact resistance measurement data must be provided.

[0064] The electrical intelligent agent calls the resistance measurement tool through the MCP server to perform a self-test. After confirming that everything is normal, it replies with message F. The content of message F is as follows: Intent: To submit evidence; Content: Contact resistance is within the normal range (R<0.1mΩ), ruling out loose connections; Attachment: Resistance measurement timing data package.

[0065] S13: Multiple rounds of questioning and causal arguments.

[0066] After the arbitration agent eliminates electrical problems using the process of elimination, it initiates a deep inquiry into the remaining possibilities.

[0067] First round of questioning (to the charging module agent): The arbitration agent sends message G to the charging module agent, asking: "Is the harmonics the root cause of the heat generation, or does the heat generation cause the harmonics?" The content of message G is as follows: Intent: To request evidence; Content: Please provide a causal analysis of harmonics and junction temperature.

[0068] First round of verification: The charging module agent calls the causal analysis tool in the MCP server to analyze historical data sequences and replies with message H: Intent: To support a viewpoint; Content: Analysis shows that temperature rise precedes harmonic distortion, and high temperature causes IGBT switching characteristics to drift. Harmonics are a result, not a cause. Attachment: Causal Analysis Report.

[0069] Second round of questioning (for the thermal management agent): The arbitration agent sends message I, asking: "Is the current temperature difference sufficient to prove that the heat dissipation system is faulty?" Message I contains the following: Intent: To request evidence; Content: Please prove that insufficient heat dissipation is sufficient to cause the current problem.

[0070] Second round of verification: The thermal management agent calls the parameter adjustment simulator in the MCP server, sets the heat dissipation capacity attenuation parameter for simulation, and replies with message J: Intent: To support a viewpoint; Content: Simulation results show that if the heat dissipation capacity decreases by 30% (e.g., due to fan aging), the junction temperature will rise by 15°C under the current load, which is consistent with the measured data.

[0071] S14: Final decision and distribution of results.

[0072] Overall Ruling: Based on the above chain of evidence (normal electrical connection + harmonics are the effect, not the cause + simulation of heat dissipation attenuation matches), the arbitration agent determines that the root cause of the fault is: insufficient heat dissipation efficiency due to aging of the heat dissipation system, and the secondary cause is: increased thermal stress on the power devices.

[0073] The arbitration agent sends the arbitration result message to all participating parties via the A2A protocol.

[0074] Send the following message to the charging module's intelligent agent: Confirm that the source of the fault is external, and suggest entering a derating mode to protect the devices.

[0075] Send to the thermal management agent: Confirm that its diagnosis has been adopted and mark it as the root cause.

[0076] Send to the electrical intelligent agent: Confirm the validity of its evidence and release it from task occupation.

[0077] In addition, it sends structured data containing the root cause to the process orchestration agent, triggering subsequent work order generation and task bidding processes.

[0078] This embodiment achieves a logical reasoning process similar to human expert consultation through multi-round structured message exchange organized by an arbitration agent. Unlike traditional rule-based fault code mapping, this process can handle faults with complex causal relationships and those spanning multiple subsystems. Furthermore, the entire debate process retains complete A2A message records, making the fault diagnosis results highly interpretable and auditable.

[0079] In another embodiment, the arbitration agent sends the arbitration result to the process orchestration agent via the A2A protocol. After receiving the arbitration result, the process orchestration agent generates an operation and maintenance work order based on the root cause of the fault.

[0080] Specifically, the process orchestration agent selects the executor based on the complexity of the maintenance work order, as follows: Simple tasks: If only parameter reset is required, assign the task directly to the default execution agent.

[0081] Complex / Multi-tasking: Involves hardware replacement (such as replacing a fan), initiates a task bidding process based on the A2A protocol, and sends the maintenance work order to the winning execution agent.

[0082] The winning intelligent agent then executes the tasks outlined in the maintenance work order and synchronizes the task status in real time with all relevant expert intelligent agents via the A2A protocol. The relevant expert intelligent agents verify the post-maintenance data via the MCP protocol and provide feedback on the verification results via the A2A protocol.

[0083] like Figure 4 The following example illustrates how to implement intelligent bidding, scheduling, and closed-loop management of preventive maintenance tasks for charging piles through the A2A protocol.

[0084] 1. Smart Proposal Stage.

[0085] S20: Routine monitoring and early warning of anomalies.

[0086] During normal operation of the charging pile, the charging module intelligent agent uses its corresponding MCP server to call the capacitor ESR calculator tool and continuously monitor the capacitor operation data obtained from the SCADA real-time database. When the charging module intelligent agent analyzes and finds that the equivalent series resistance (ESR) value of a charging module's capacitor is continuously increasing and exceeds the preset preventive maintenance threshold, it determines that the capacitor has an aging trend, which may lead to future failure.

[0087] S21: Based on the warning of excessive ESR, the charging module agent automatically generates a structured preventive maintenance proposal. This proposal is sent to the process orchestration agent via the A2A protocol. The proposal includes information such as: the faulty component (the capacitor of charging module X), the recommended operation (replacing the capacitor), the urgency level (medium), the estimated time, and the type of spare parts required.

[0088] 2. Task assignment decision-making and tendering announcement stage.

[0089] S22: After receiving the proposal from the module agent, the process orchestration agent first assesses the complexity of the task, resource requirements, and whether it involves multiple charging piles or subsystems to determine whether bidding is required to confirm the executor. If not, proceed to S23; if yes, proceed to S24.

[0090] S23: The task is determined to be a simple task (e.g., only requiring remote parameter reset). The process orchestration agent directly assigns it to the preset default execution agent (e.g., remote control agent) via the A2A protocol, and skips the bidding process, directly entering the execution and monitoring stage of step S30 below.

[0091] S24: If the task is determined to be a complex task (e.g., it requires on-site hardware replacement and may involve batch maintenance of multiple charging piles of the same type), the process orchestration agent will initiate the task bidding process.

[0092] S25: The process orchestration agent integrates all pending preventative maintenance proposals, particularly merging similar maintenance needs for multiple charging piles within the same charging station. Subsequently, the process orchestration agent broadcasts a structured tender message via the A2A protocol to the registered group of executor agents (including field maintenance robot agents, external service provider agents, internal technical team agents, etc.). This tender message details the task (e.g., the quantity, location, and completion deadline of capacitors to be replaced), quality constraints, and requirements for the tender proposals (e.g., cost budget, estimated completion time, and qualifications of tools / personnel used).

[0093] 3. Bidding and evaluation stage.

[0094] S26: Bidding by Executing Agents. After receiving the bidding message, each executing agent sends a structured bidding message to the process orchestration agent via the A2A protocol, based on its own capabilities, resource availability, and cost model.

[0095] Implementing agent A (e.g., field maintenance robot agent): Submit a tender message including "Solution: Automatic robot replacement", "Cost: Low (consumables and power only)", and "Time: Completed within 1 hour".

[0096] Executor B (e.g., external service provider): Submits a tender message including "Solution: Dispatch a professional engineer to replace on-site", "Cost: Medium (including labor costs)", and "Time: Completed within 4 hours (including travel)".

[0097] The implementing agent C (e.g., an internal technical team agent): submits a tender message containing "Solution: Internal engineer scheduling", "Cost: Medium (internal man-hours)", and "Time: Completed the next day".

[0098] All bidding messages are aggregated into the built-in evaluation module of the process orchestration agent.

[0099] 4. Evaluation and decision-making stage and commitment confirmation stage.

[0100] S27: The process orchestration agent evaluates all received bids according to a pre-set evaluation strategy (e.g., a weighted scoring model that comprehensively considers factors such as cost, efficiency, quality, and urgency) to determine if any bids are qualified. If not, it means that no bids are qualified after evaluation, or all bids do not meet the requirements. The process orchestration agent will then determine that the bidding process has failed and choose to reissue the tender or adjust the tender requirements, returning to the tender issuance stage in S25. If yes, proceed to step S28 below.

[0101] S28: The process orchestration agent ultimately selects the optimal bidding solution (e.g., selecting the robot solution of executor A because it has the lowest cost and the highest efficiency).

[0102] S29: Task Commitment and Notification. The process orchestration agent sends a task commitment request to the successful executor agent A via the A2A protocol. Upon receiving the request, executor agent A replies with a "commitment confirmation" message via the A2A protocol, thereby locking in execution resources and time. Simultaneously, the process orchestration agent also sends A2A notification messages to the unsuccessful executors B and C, informing them of the bidding results.

[0103] 5. Execution and monitoring phase.

[0104] S30: Task Execution and Status Synchronization. After receiving the A2A instruction, the successful executor, Agent A (the robot), proceeds to the designated charging station and controls the charging station equipment via its MCP interface (e.g., using the MCP server to call a robotic arm to control a tool for capacitor replacement). During task execution, Agent A synchronizes the task status (e.g., "Arrived on site," "Replacing capacitor," "Replacement complete," etc.) in real time with the process orchestration agent and relevant expert agents via the A2A protocol. Upon task completion, Agent A sends a task completion confirmation and relevant data feedback.

[0105] 6. Verification and knowledge closure stage.

[0106] S31: Repair Effectiveness Verification. After the task is completed, the charging module's intelligent agent again calls the capacitor ESR calculator tool through the MCP server to monitor the ESR value of the replaced capacitor. If the ESR value returns to the normal range, the repair is verified as successful.

[0107] S32: Knowledge Updates and Model Optimization. All data from the entire process of this preventative maintenance task, from proposal and bidding to execution and verification (including A2A message flow, MCP call records, maintenance results, etc.), is automatically archived into the knowledge base by the system. Predictive models of the charging module agent and other relevant expert agents (e.g., capacitor life prediction models) are incrementally trained using training tools managed by the MCP server to improve the accuracy of future predictions and achieve system self-evolution.

[0108] This embodiment achieves dynamic optimization of operation and maintenance resources through an A2A protocol-driven task bidding mechanism. It intelligently selects the optimal executor based on task requirements and resource availability, thereby improving operation and maintenance efficiency and reducing operating costs. Simultaneously, complete A2A message flow records ensure the auditability of the decision-making process, while the knowledge loop enables the system to learn and evolve from each operation and maintenance experience, ultimately achieving smarter and more proactive preventative maintenance.

[0109] In summary, this invention discloses a charging pile operation and maintenance method based on multi-agent collaboration, such as... Figure 5 Its workflow is as follows: S1. System initialization and routine monitoring.

[0110] After the system powers on, the MCP server is first configured to encapsulate heterogeneous system data and functions, and expert agents are then configured. Subsequently, each expert agent and its corresponding MCP server start up. The agents subscribe to the data streams they are interested in through the MCP server, enter a normalized monitoring state, and continuously use MCP tools to analyze the data.

[0111] S2. Abnormality perception and preliminary diagnosis.

[0112] During routine monitoring, any expert agent (e.g., the charging module agent) detects an abnormal event (e.g., output power fluctuation or excessive capacitor ESR) using the MCP tool and generates an abnormal event. At this point, the system determines whether it can independently diagnose the issue. If the problem can be diagnosed independently (for a single, simple task), the process will jump to the maintenance work order generation stage in step S4 below.

[0113] If external collaboration is required (for complex / multi-component issues), the process will proceed to the collaborative diagnosis and arbitration phase in step S3 below.

[0114] S3. Collaborative diagnosis and arbitration.

[0115] The expert agent that detects the anomaly conducts bilateral negotiations with relevant expert agents via the A2A protocol. If no agreement can be reached or the fault is complex, an arbitration request message is sent to the arbitration agent.

[0116] It should be noted that if the expert agent that senses the anomaly reaches a consensus through bilateral negotiation with the relevant expert agents, that is, the cause of the fault is confirmed, the process will directly proceed to the arbitration process and enter step S4 below.

[0117] After receiving the request, the arbitration agent organizes all relevant expert agents through the A2A protocol to conduct multiple rounds of structured message exchange in order to jointly diagnose and ultimately determine the root cause of the failure.

[0118] S4. Maintenance Work Order Generation Whether the task is directly classified as a simple task in the S2 phase, the root cause is determined by the arbitration agent in the S3 phase, or the root cause of the fault is directly determined without going through the arbitration agent in the S3 phase, the system will generate maintenance work orders based on fault information or preventive maintenance proposals through the process orchestration agent.

[0119] S5. Task scheduling and resource allocation.

[0120] After receiving an operation and maintenance work order, the process orchestration agent makes decisions based on the task's complexity, resource requirements, and other factors. Simple task assignment: If it is a simple task, the process orchestration agent will directly assign it to the preset default execution agent through the A2A protocol.

[0121] Complex Task Bidding: For complex tasks, the process orchestration agent will initiate a task bidding process based on the A2A protocol. Each executor agent submits a bid message through the A2A protocol. After evaluating the bids, the process orchestration agent sends a task commitment request and confirmation to the winning bidder through the A2A protocol.

[0122] S6. Task execution and state synchronization.

[0123] After receiving the A2A instruction, the executing agent (whether directly assigned or selected through bidding) proceeds to the designated location to perform the task, and synchronizes the task status in real time with the process orchestration agent and relevant expert agents via the A2A protocol. Upon completion of the task, the executing agent sends a task completion confirmation and relevant data feedback. The relevant expert agents verify the post-repair data via the MCP protocol and provide feedback on the verification results via the A2A protocol.

[0124] S7. Knowledge Loop and System Evolution.

[0125] The complete case study of this maintenance task (including A2A message flow, MCP call records, maintenance work orders, maintenance results, etc.) was automatically structured and archived into the knowledge base by the system. The predictive models of relevant expert agents were incrementally trained and optimized through training tools managed by the MCP server to improve the accuracy of future fault prediction and diagnostic efficiency.

[0126] With this, the operation and maintenance process is now complete, the system has resumed normal monitoring, and can continue to learn and evolve.

[0127] The present invention also discloses a charging pile operation and maintenance system based on multi-agent collaboration. This operation and maintenance system works based on the charging pile operation and maintenance method in the above embodiments to perform fault maintenance on the charging pile.

[0128] This invention also discloses another charging pile operation and maintenance system, which includes one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors. The programs include instructions for performing the charging pile operation and maintenance method as described above. The processor may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, used to execute relevant programs to implement the functions required by the modules in the charging pile operation and maintenance system of this application embodiment, or to execute the charging pile operation and maintenance method of this application method embodiment.

[0129] This invention also discloses a computer-readable storage medium comprising a computer program that can be executed by a processor to perform the charging pile operation and maintenance method described above. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be read-only memory (ROM), random access memory (RAM), or magnetic media, such as floppy disks, hard disks, magnetic tapes, magnetic disks, or optical media, such as digital versatile discs (DVDs), or semiconductor media, such as solid-state disks (SSDs).

[0130] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned charging pile operation and maintenance method.

[0131] The above-disclosed embodiments are merely preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A charging pile operation and maintenance method based on multi-agent collaboration, characterized in that, include: Configure multiple Application Model Context Protocol (MCP) servers corresponding to each functional subsystem in the charging pile. The MCP servers are used to encapsulate the data interfaces and functions of the back-end heterogeneous systems related to each functional subsystem into resources and tools that conform to the MCP protocol standard. Multiple expert agents are configured, each corresponding to a specific MCP server. The expert agents are configured to invoke the resources and tools provided by the MCP server through the MCP protocol to obtain standardized data and perform functional subsystem-level anomaly detection and preliminary diagnosis. When any of the aforementioned expert agents perceives an anomaly and determines that external collaboration is required, it conducts bilateral negotiations with other expert agents associated with the anomaly through the inter-agent communication protocol A2A, and sends an arbitration request message to the arbitration agent when it is determined that multi-party consultation is required. After receiving the arbitration request message, the arbitration agent organizes all relevant expert agents through the A2A protocol to conduct multiple rounds of structured message exchange in order to jointly diagnose and determine the root cause of the fault. A maintenance work order is generated based on the root cause of the fault.

2. The charging pile operation and maintenance method according to claim 1, characterized in that, The expert agent subscribes to the data streams it is interested in through the MCP server for routine monitoring, and calls the tools provided by the MCP server to analyze the data streams in order to detect anomalies and generate functional subsystem-level abnormal events.

3. The charging pile operation and maintenance method according to claim 1, characterized in that, The two expert agents achieve bilateral negotiation by sending structured A2A messages, which include at least one of intent, context, content, and constraints.

4. The charging pile operation and maintenance method according to claim 1, characterized in that, After receiving the bilateral negotiation message, other expert agents associated with the anomaly invoke tools and query resources through their corresponding MCP servers to collect evidence, and reply with analysis results through the A2A protocol. The analysis results include at least one of intent, content, and suggestions.

5. The charging pile operation and maintenance method according to claim 1, characterized in that, The arbitration agent organizes all relevant expert agents to conduct multi-round structured message exchanges, including sending one or more of the following: invitation to join message, request for evidence message, question message, receiving evidence provision message, submit analysis message, and submit simulation message.

6. The charging pile operation and maintenance method according to claim 1, characterized in that, The arbitration agent sends the arbitration result to the process orchestration agent via the A2A protocol; after receiving the arbitration result, the process orchestration agent generates the maintenance work order based on the root cause of the fault.

7. The charging pile operation and maintenance method according to claim 6, characterized in that, The process orchestration agent sends the maintenance work order to the default execution agent based on the complexity of the maintenance work order, or initiates a task bidding process based on the A2A protocol to send the maintenance work order to the winning execution agent.

8. The charging pile operation and maintenance method according to claim 7, characterized in that, The winning execution agent executes the tasks in the maintenance work order and synchronizes the task status with all relevant expert agents in real time through the A2A protocol; the relevant expert agents verify the post-maintenance data through the MCP protocol and provide feedback on the verification results through the A2A protocol.

9. The charging pile operation and maintenance method according to claim 1, characterized in that, After the maintenance work order is closed, the A2A message stream and MCP call records generated during this maintenance are stored in the knowledge base; the prediction model of the relevant expert agent is incrementally trained and optimized through the tools managed by the MCP server.

10. A charging pile operation and maintenance system based on multi-agent collaboration, characterized in that, The operation and maintenance system operates based on the charging pile operation and maintenance method according to any one of claims 1 to 9, in order to perform fault maintenance on the charging pile.

11. A charging pile operation and maintenance system, characterized in that, include: One or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the charging pile operation and maintenance method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, It includes a computer program that can be executed by a processor to perform the charging pile operation and maintenance method as described in any one of claims 1 to 9.

Images