Hermes-based virtual power plant multi-agent scheduling method and device

Abstract

The application discloses a kind of virtual power plant multi-agent scheduling method and device based on Hermes, it is related to virtual power plant distributed resource scheduling optimization technical field, the method includes obtaining virtual power plant heterogeneous resource data, constructs unified resource state vector and encapsulation;Multi-agent scheduling cluster is constructed, external scheduling demand is parsed as scheduling task contract and is distributed, each agent generates ability offer message, feasible region boundary and risk prompt, obtain scheduling negotiation result after completing scheduling negotiation;Multi-objective safety constraint scheduling optimization model is constructed and virtual power plant decomposed scheduling plan is generated;Control instruction is generated and is issued, equipment real-time execution data is collected, when preset trigger condition is satisfied, start re-negotiation optimization;Finally, chain hash storage evidence method is used to realize scheduling whole process traceable audit.The application can improve the collaborative efficiency of heterogeneous resources, reduce scheduling delay, and realize the traceable audit of the whole link process of virtual power plant scheduling.

Description

Technical Field

[0001] This application relates to the field of distributed resource scheduling and optimization technology for virtual power plants, and in particular to a multi-agent scheduling method and apparatus for virtual power plants based on Hermes-like architecture. Background Technology

[0002] In the new power system, distributed photovoltaic, decentralized wind power, user energy storage, charging piles, and industrial / building flexible loads are massively integrated, making virtual power plants (VPPs) the core carrier for aggregating decentralized resources to participate in grid dispatch and electricity market transactions. Existing VPP dispatch schemes mainly include centralized optimization dispatch schemes, rule-driven or hierarchical control schemes, and general multi-agent dispatch schemes. However, these schemes generally suffer from shortcomings such as the disconnect between resource status updates and dispatch solutions, insufficient coupling between grid constraints and market / carbon constraints, lack of automatic correction loops in edge execution feedback, and reliance on manual intervention for abnormal operating conditions. They cannot simultaneously solve technical pain points such as low efficiency of heterogeneous resource collaboration, high dispatch latency, and lack of end-to-end traceability.

[0003] In summary, there is an urgent need to provide a multi-agent scheduling method for virtual power plants that offers high heterogeneous resource collaboration efficiency, low scheduling latency, and full-link traceability and auditability for virtual power plant scheduling. This would simultaneously address the problems of low heterogeneous resource collaboration efficiency, high scheduling latency, and lack of full-link traceability in existing virtual power plant scheduling schemes. Summary of the Invention

[0004] The purpose of this application is to provide a multi-agent scheduling method and device for virtual power plants based on Hermes, which can improve the efficiency of heterogeneous resource collaboration, reduce scheduling delay, and realize full-link traceability and auditing of virtual power plant scheduling.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a multi-agent scheduling method for virtual power plants based on a Hermes-like framework. The Hermes-like framework is a messenger-based multi-agent collaborative messaging mechanism for energy scheduling scenarios, possessing full-link capabilities including unified message envelope encapsulation, standardized scheduling contract distribution, distributed negotiation among multiple agents, resource status version synchronization, message receipt confirmation, and closed-loop processing of abnormal events. The method includes the following steps: Acquire heterogeneous resource data of virtual power plants.

[0006] Based on the heterogeneous resource data of the virtual power plant, a unified resource state vector is constructed and encapsulated as a Hermes-like scheduling message envelope.

[0007] A multi-agent scheduling cluster is constructed based on the Hermes-like scheduling message envelope. External scheduling requirements are parsed into scheduling task contracts and distributed to each agent in the multi-agent scheduling cluster. Each agent generates capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustability, and local constraints. The scheduling negotiation is completed based on the messenger-style multi-agent collaborative message mechanism to obtain the scheduling negotiation result.

[0008] Based on the capability quotation message, feasible domain boundary and risk warning, a multi-objective security constraint scheduling optimization model is constructed and scheduling conflict arbitration priority scoring is performed to generate a virtual power plant decomposition scheduling plan.

[0009] Control commands are generated based on the decomposed scheduling plan of the virtual power plant and distributed to each edge execution device through the edge execution agent in the multi-agent scheduling cluster. Real-time execution data of each edge execution device is collected, and local rolling renegotiation optimization is automatically started when the preset trigger conditions are met to obtain the rescheduling negotiation result.

[0010] Each of the scheduling task contract, the capability quotation message, the scheduling negotiation result, the scheduling conflict arbitration priority score, the control command, the device real-time execution data, and the rescheduling negotiation result is assigned a unique identifier, a timestamp, a digital signature, and a status version number. A chain hash notarization method is used to achieve full-process traceability and auditing of the scheduling process.

[0011] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the Hermes-based virtual power plant multi-agent scheduling method.

[0012] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the described Hermes-based virtual power plant multi-agent scheduling method.

[0013] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the Hermes-based virtual power plant multi-agent scheduling method.

[0014] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a Hermes-like multi-agent scheduling method and apparatus for virtual power plants. First, it constructs a unified resource state vector and standardizes the encapsulation of various heterogeneous resource data from virtual power plants using Hermes-like scheduling message envelopes. Relying on a Hermes-like messenger-style multi-agent collaborative messaging mechanism with standardized message interaction capabilities, it uniformly carries all interactive data throughout the scheduling process, eliminating data expression differences among different types of distributed resources and achieving unified interoperability of data from various heterogeneous resources in the virtual power plant, providing a standardized data interaction foundation for multi-agent collaborative scheduling. Second, it establishes a multi-agent scheduling cluster with clearly defined roles, relying on the Hermes-like messenger-style multi-agent collaborative messaging mechanism to complete contract distribution, distributed negotiation, state version synchronization, acknowledgment confirmation, and anomaly closed-loop processing. Each agent outputs capability quotation messages, feasible domain boundaries, and risk warnings based only on its own local constraints. The distributed negotiation mode replaces single centralized computation, reducing the centralized computing power pressure and scheduling latency of unified scheduling, thereby improving the efficiency of heterogeneous resource collaboration. Furthermore, by integrating the capability quotes, feasible domain boundaries, and risk warnings reported by various agents, a multi-objective safety constraint scheduling optimization model is constructed. Limited adjustment resources are quantified and allocated using a scheduling conflict arbitration priority scoring method. This allows for the orderly handling of conflict scenarios where multiple tasks compete for the same adjustment resource, achieving an optimal decomposed scheduling plan output that balances multiple scheduling objectives. Moreover, edge-execution agents complete the issuance of control commands and the collection of real-time data from devices. When preset trigger conditions are met, local rolling renegotiation optimization is automatically performed. This allows for the regeneration of scheduling schemes for periods without execution, eliminating the need for repeated calculations across all scheduling cycles. This achieves automated closed-loop correction of scheduling deviations and enhances the adaptive capability of scheduling operations. In addition, by attaching unique identifiers, timestamps, digital signatures, and status version numbers to core business data throughout the entire scheduling process, and employing a chain-hash notarization method, the entire scheduling process is traceable and auditable. This enables complete tracing of the decision-making basis and execution process for the entire scheduling negotiation, optimization arbitration, command issuance, device execution, and re-optimization process, ensuring that the entire scheduling process is auditable, verifiable, and tamper-proof. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 An application environment diagram of a virtual power plant multi-agent scheduling method based on Hermes class provided in an embodiment of this application; Figure 2A schematic diagram of the overall process of a multi-agent scheduling method for a virtual power plant based on a Hermes class, provided as an embodiment of this application; Figure 3 A schematic diagram of the core steps of a virtual power plant multi-agent scheduling method based on Hermes class provided in an embodiment of this application; Figure 4 A schematic diagram of a Hermes-like multi-agent cooperative scheduling architecture provided in an embodiment of this application; Figure 5 A schematic diagram of a multi-objective security constraint scheduling optimization and conflict arbitration process provided in an embodiment of this application; Figure 6 A schematic diagram of a multi-agent scheduling system architecture for a virtual power plant based on a Hermes-like class provided in an embodiment of this application; Figure 7 A schematic diagram of the structure of a virtual power plant multi-agent scheduling device based on Hermes class, provided as an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] Terminology Explanation: Hermes-like: A messenger-style multi-agent collaborative messaging mechanism with full-link capabilities such as unified message envelope encapsulation, standardized scheduling contract distribution, multi-agent distributed negotiation, resource state version synchronization, message receipt confirmation, closed-loop handling of abnormal events, and traceable recording.

[0019] Scheduling task contract: A structured scheduling task description transformed from external scheduling requirements, which includes at least the task number, target power curve, response time window, capacity boundary, security constraints, settlement rules, penalty for breach of contract, and anomaly degradation strategy.

[0020] Unified Resource State Vector: A structured data set used to describe the adjustability, operating status, economic parameters, carbon emission parameters, communication status, and security constraints of heterogeneous resources within a virtual power plant at the same scheduling time granularity.

[0021] Hermes-like scheduling message envelope: A data structure that uniformly encapsulates task, quote, constraint, feedback, and exception messages. It includes message identifier, sender, receiver or subscribed topic, message type, timestamp, scheduling time window, message payload, constraint set, confidence level, priority, status version number, acknowledgment identifier, digital signature, and exception handling flag.

[0022] Decompositional scheduling plan: A power execution plan at the equipment level or cluster level formed by decomposing the overall scheduling target of the virtual power plant into different resource devices or resource clusters.

[0023] With the continuous increase in installed capacity of new energy sources and the rapid growth of flexible resources on the user side, the traditional power system is shifting from a centralized, unidirectional power supply model to a collaborative and interactive model involving power generation, grid, load, and storage. Virtual power plants, through information and communication technologies and aggregation control technologies, aggregate a large number of dispersed resources such as distributed photovoltaic power, wind power, energy storage, charging piles, adjustable loads, building air conditioning, industrial flexible loads, and demand response terminals into a dispatchable whole, participating in electricity market transactions, ancillary services, peak shaving and valley filling, demand response, and new energy consumption.

[0024] Existing virtual power plant dispatching technologies mainly include the following types of solutions: The first category is centralized optimization scheduling schemes. These schemes typically involve a virtual power plant central control platform that collects the status of all resources and generates an overall scheduling plan using mixed-integer programming, linear programming, dynamic programming, or heuristic algorithms. This approach is effective when resource scale is small, communication is stable, and constraints are few. However, as the number of connected resources increases, the scheduling model's dimensionality expands rapidly, significantly increasing solution time. Furthermore, the centralized platform needs to grasp the complete status of all resources, leading to high communication pressure and potential single-point bottlenecks. In high-frequency real-time markets, rapid ancillary services, and scenarios involving sudden changes in power grid constraints, centralized scheduling struggles to generate executable plans in a timely manner.

[0025] The second category is rule-driven or hierarchical control schemes. These schemes typically trigger actions such as energy storage charging / discharging, load shedding, or charging pile power limiting based on preset thresholds, time periods, electricity price ranges, or load levels. While simple to implement, these schemes lack the comprehensive optimization capabilities to address resource uncertainty, user comfort, grid safety boundaries, market returns, and carbon constraints. When renewable energy output, load demand, and market prices fluctuate simultaneously, fixed rules are prone to over-adjustment, insufficient response, or excessive equipment lifespan degradation.

[0026] The third category is general multi-agent scheduling schemes. These schemes abstract distributed resources into multiple agents, achieving resource scheduling through local negotiation or distributed optimization. However, existing multi-agent schemes typically only focus on algorithm-level collaboration, lacking standardized message structures, task contract mechanisms, scheduling receipt mechanisms, and execution closed-loop mechanisms adapted to virtual power plant operations. This leads to inconsistencies in state semantics between different agents, insufficient constraint propagation, opaque conflict arbitration, and difficulty in auditing the scheduling process, making it difficult to meet the power system's requirements for security, real-time performance, interpretability, and traceability.

[0027] Furthermore, existing virtual power plant systems generally suffer from problems such as the disconnect between resource status updates and scheduling solutions, insufficient coupling between market signals and distribution network security constraints, difficulty in dynamically expressing user-side response intentions, and insufficient utilization of edge device execution feedback. When resource execution deviations, communication anomalies, or grid over-limit alarms occur, the system often requires manual intervention or a recalculation of the global calculations, making it difficult to achieve rapid event-driven self-correction.

[0028] In summary, existing technologies cannot simultaneously solve the technical problems of low efficiency in heterogeneous resource collaboration in virtual power plants, high latency in centralized scheduling, insufficient cross-agent constraint transmission, opaque scheduling conflict handling, insufficient execution feedback loop, and weak full-process auditing capabilities. There is an urgent need for a new scheduling architecture and method for virtual power plants that has a standardized message collaboration mechanism and multi-agent closed-loop scheduling capabilities.

[0029] The purpose of this application is to provide a Hermes-based multi-agent scheduling method for virtual power plants, applicable to day-ahead scheduling, real-time rolling optimization, power auxiliary services, demand response, local renewable energy consumption, and carbon collaborative management scenarios in park, regional, and city-level virtual power plants. Through a Hermes-like message bus, scheduling task contracts, unified resource state vectors, multi-agent negotiation, multi-objective security constraint optimization, edge-cloud collaborative closed-loop execution, and trusted auditing mechanisms, this method addresses the technical problems of low heterogeneous resource collaboration efficiency, high centralized scheduling latency, insufficient constraint propagation, difficulty in arbitrating scheduling conflicts, difficulty in closing the execution feedback loop, and weak full-process traceability in virtual power plants. While ensuring grid security and user-side response constraints, it reduces scheduling latency, improves heterogeneous resource collaboration efficiency, and enhances the real-time scheduling capability, renewable energy consumption capability, market revenue capability, and full-link traceability and auditability of virtual power plants.

[0030] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0031] The Hermes-based virtual power plant multi-agent scheduling method provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on other servers. Terminal 102 can send heterogeneous resource data of a virtual power plant to server 104. After receiving the heterogeneous resource data, server 104 constructs a unified resource state vector based on the data and encapsulates it into a Hermes-like scheduling message envelope. Based on the Hermes-like scheduling message envelope, a multi-agent scheduling cluster is constructed, and external scheduling requirements are parsed into scheduling task contracts and distributed to each agent in the multi-agent scheduling cluster. This allows each agent to generate capability offer messages, feasible domain boundaries, and risk warnings based on its own resource state, adjustable capabilities, and local constraints. Scheduling negotiation is then completed based on the messenger-style multi-agent collaborative messaging mechanism to obtain the scheduling negotiation result. Based on the capability offer messages, feasible domain boundaries, and risk warnings, a multi-objective scheduling cluster is constructed. A safety-constrained scheduling optimization model is used to perform scheduling conflict arbitration priority scoring, generating a virtual power plant decomposed scheduling plan. Based on this plan, control commands are generated and distributed to each edge execution device through edge execution agents in the multi-agent scheduling cluster. Real-time execution data of each edge execution device is collected. When preset trigger conditions are met, local rolling renegotiation optimization is automatically initiated to obtain a rescheduling negotiation result. Unique identifiers, timestamps, digital signatures, and status version numbers are added to the scheduling task contract, the capability quotation message, the scheduling negotiation result, the scheduling conflict arbitration priority score, the control command, the device real-time execution data, and the rescheduling negotiation result. A chain-hash notarization method is used to achieve full-process traceability and auditing of the scheduling process. Server 104 can feed back the obtained scheduling task contract, capability quotation message, scheduling negotiation result, scheduling conflict arbitration priority score, control command, device real-time execution data, rescheduling negotiation result, and audit index log records to terminal 102. In addition, in some embodiments, the Hermes-based virtual power plant multi-agent scheduling method can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly perform virtual power plant multi-agent scheduling processing on the heterogeneous resource data of the virtual power plant, or the server 104 can obtain the heterogeneous resource data of the virtual power plant from the data storage system and perform virtual power plant multi-agent scheduling processing on the heterogeneous resource data of the virtual power plant.

[0032] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, and IoT devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0033] In one exemplary embodiment, such as Figure 2 As shown, a multi-agent scheduling method for a virtual power plant based on the Hermes class is provided. This method is executed by computer equipment, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the following steps are included: S1: Obtain heterogeneous resource data of virtual power plant.

[0034] S2: Based on the heterogeneous resource data of the virtual power plant, construct a unified resource state vector and encapsulate it into a Hermes-like scheduling message envelope.

[0035] S3: Construct a multi-agent scheduling cluster based on the Hermes-like scheduling message envelope, and parse the external scheduling requirements into scheduling task contracts and distribute them to each agent in the multi-agent scheduling cluster. This enables each agent to generate capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustable capabilities, and local constraints. The agent then completes scheduling negotiation based on the messenger-style multi-agent collaborative message mechanism to obtain the scheduling negotiation result.

[0036] S4: Based on the capability quotation message, feasible domain boundary and risk warning, construct a multi-objective security constraint scheduling optimization model and perform scheduling conflict arbitration priority scoring to generate a virtual power plant decomposition scheduling plan.

[0037] S5: Based on the decomposed scheduling plan of the virtual power plant, control instructions are generated and sent to each edge execution device through the edge execution agent in the multi-agent scheduling cluster. Real-time execution data of each edge execution device is collected. When the preset triggering conditions are met, local rolling renegotiation optimization is automatically started to obtain the rescheduling negotiation result.

[0038] S6: Each of the following is assigned a unique identifier, timestamp, digital signature, and status version number: scheduling task contract, capability quotation message, scheduling negotiation result, scheduling conflict arbitration priority score, control command, real-time device execution data, and rescheduling negotiation result. A chain hash notarization method is used to achieve full-process traceability and auditing of scheduling.

[0039] Implementing steps S1 to S6 above, relying on a Hermes-like multi-agent collaborative messaging mechanism with unified message envelope encapsulation capabilities, standardized scheduling contract distribution capabilities, multi-agent distributed negotiation capabilities, resource status version synchronization capabilities, message receipt confirmation capabilities, and abnormal event closed-loop processing capabilities, a unified resource status vector is first constructed for heterogeneous resource data of virtual power plants, and messages are encapsulated in a standardized manner. A multi-agent scheduling cluster is then built to complete scheduling contract distribution and distributed negotiation. Scheduling resource conflicts are resolved by relying on a multi-objective security constraint scheduling optimization model combined with scheduling conflict arbitration priority scoring. Then, instructions are issued by edge execution agents, real-time execution data is collected from equipment, and local rolling renegotiation is automatically triggered to correct operational deviations. At the same time, unique identifiers, timestamps, digital signatures, and version numbers are added to various business data in the entire scheduling process and stored in a chain hash. This not only unifies the heterogeneous resource data interaction standard, reduces the computing power pressure of centralized scheduling, ensures the real-time effectiveness of scheduling data, orderly handles multi-task resource conflicts, and adaptively corrects scheduling operational deviations, but also improves the efficiency of heterogeneous resource collaboration, reduces scheduling latency, and achieves tamper-proof, complete, traceable, and auditable scheduling processes throughout the entire process.

[0040] As an optional implementation, in step S1, the heterogeneous resources of the virtual power plant include one or more of the following: distributed photovoltaic, distributed wind power, gas-fired combined cooling, heating and power units, electrochemical energy storage, user-side energy storage, charging pile clusters, interruptible loads, transferable loads, temperature-controlled loads, industrial flexible loads, building air conditioning loads, distribution transformers, feeders, switchgear, reactive power compensation devices and electricity metering terminals.

[0041] As an optional implementation, in step S2, the unified resource state vector includes: unique resource identifier, resource type, grid-connected node, maximum power generation or adjustable power, minimum power generation or adjustable power, ramp-up rate, ramp-down rate, sustainable response time, energy storage state of charge, state confidence, prediction uncertainty, operating cost, carbon emission factor, user comfort constraints, equipment lifespan loss parameters, communication availability, and security constraint labels.

[0042] As an optional implementation, in step S2, the Hermes-like scheduling message envelope includes: a message identifier, a sending agent, a receiving agent or subscribed topic, a message type, a timestamp, a scheduling time window, a message payload, a constraint set, a confidence level, a priority, a status version number, an acknowledgment identifier, a digital signature, and an exception handling flag; the message type includes task release messages, status update messages, capability quote messages, constraint propagation messages, conflict alarm messages, scheduling commitment messages, control instruction messages, execution feedback messages, and re-optimization trigger messages.

[0043] As an optional implementation, in step S3, the scheduling task contract includes a task number, scheduling source, target power curve, response start time, response duration, minimum response capacity, maximum response capacity, reserve capacity requirement, voltage safety constraint, power flow safety constraint, market settlement price, default penalty, carbon emission constraint, resource type preference, response confirmation deadline, and anomaly degradation strategy.

[0044] As an optional implementation, in step S3, a multi-agent scheduling cluster is constructed based on the Hermes-like scheduling message envelope, and external scheduling requirements are parsed into scheduling task contracts and distributed to each agent in the multi-agent scheduling cluster. This allows each agent to generate capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustability, and local constraints, and to complete scheduling negotiation based on the messenger-style multi-agent cooperative messaging mechanism to obtain the scheduling negotiation result. Specifically, this includes the following steps: S31: Based on the Hermes-like scheduling message envelope, and combined with the resource type, scheduling objective, and operational constraints of the virtual power plant, construct a multi-agent scheduling cluster.

[0045] S32: Based on the multi-agent scheduling cluster, receive external scheduling requests from the power grid, the electricity market, and the demand response side.

[0046] S33: Parse the external scheduling requirements to obtain the scheduling task contract; the scheduling task contract includes: power requirements, time scale, constraints, response priority and settlement rules.

[0047] S34: Utilizing the standardized scheduling contract distribution capability of the messenger-based multi-agent collaborative messaging mechanism, the scheduling task contract is distributed to each agent in the multi-agent scheduling cluster, so that each agent generates corresponding capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustable capabilities, and local constraints.

[0048] S35: Based on the capability quote messages, feasible domain boundaries, and risk warnings corresponding to each agent, the multi-agent distributed negotiation capability, resource state version synchronization capability, message receipt confirmation capability, and abnormal event closed-loop processing capability of the messenger-type multi-agent cooperative message mechanism are used to conduct scheduling negotiation among the agents and obtain the scheduling negotiation result.

[0049] As an optional implementation, in step S3, the multi-agent scheduling cluster includes: a resource agent, a prediction agent, a market agent, a grid security agent, a carbon constraint agent, a coordination agent, a risk control agent, and an edge execution agent. Specifically, the resource agent maintains the unified resource state vector for the heterogeneous resources of the corresponding virtual power plant, and generates the capacity quotation message, feasible domain boundary, and risk warning based on the scheduling task contract, its own resource state, adjustable capabilities, and local constraints; the prediction agent generates new energy output prediction results, load prediction results, charging demand prediction results, and response availability prediction results based on meteorological data, historical output data, historical load data, and user behavior data; the market agent analyzes the day-ahead market, real-time market, ancillary service market, and demand response subsidy signals; the grid security agent generates distribution network power flow constraints, voltage over-limit constraints, line capacity constraints, and reserve capacity constraints; the carbon constraint agent... The resource agent is used to calculate the carbon emission intensity and carbon constraint boundary corresponding to resource calls; the coordination agent is used to parse the external scheduling requirements to obtain the scheduling task contract and distribute it to each resource agent, summarize the capability quotation messages, feasible domain boundaries and risk warnings fed back by each resource agent and verify the status version number, construct the multi-objective security constraint scheduling optimization model and perform scheduling conflict arbitration priority scoring on each resource agent to generate a virtual power plant decomposed scheduling plan; the risk control agent is used to identify communication anomalies, equipment anomalies, response defaults and scheduling infeasibility risks; the edge execution agent is used to control the legality verification and issuance of control commands, control edge execution devices and collect real-time execution data of devices.

[0050] As an optional implementation, after receiving the capability quotation message from the resource agent, the coordinating agent compares the status version number in the capability quotation message with the latest status version number in the resource status cache. If the status version number matches the latest status version number, the capability quotation message is included in the candidate resource combination. Otherwise, the capability quotation message is discarded, and a requote request is sent to the corresponding resource agent to avoid using expired status to generate an unexecutable virtual power plant decomposition scheduling plan.

[0051] This application embodiment uses a resource status version synchronization mechanism to filter invalid capability quotes generated based on expired resource status, avoids generating unexecutable scheduling schemes based on expired resource information, ensures that resource status data is valid in real time during scheduling negotiation, and improves the executability of scheduling schemes.

[0052] As an optional implementation, in step S4, the constraints of the multi-objective security constraint scheduling optimization model include: virtual power plant power balance constraints, resource output upper and lower limit constraints, energy storage state of charge constraints, charge and discharge mutual exclusion constraints, ramp rate constraints, user-side comfort constraints, load interruption count constraints, distribution network power flow constraints, node voltage constraints, line capacity constraints, reserve capacity constraints, communication delay constraints, and market declaration boundary constraints. Among these, the supply and demand deviation penalty is calculated based on the sum of squares of the deviations between the target power curve and the planned power curve within the scheduling time window; the power grid security risk index is calculated by weighting the node voltage exceedance limit, line load rate exceedance limit, and main transformer load rate exceedance limit; and equipment life loss is calculated based on charge and discharge depth, cycle count, and life loss coefficient.

[0053] As an optional implementation, in step S5, the preset triggering conditions include at least one of the following: the deviation between the actual executed power and the planned power is greater than a preset deviation threshold; the prediction error of new energy output is greater than a preset prediction error threshold; the load prediction error is greater than a preset load error threshold; the energy storage state of charge is less than or greater than a preset safety boundary; the distribution network node voltage or line current triggers an over-limit alarm; the real-time market price fluctuation is greater than a preset price threshold; the resource agent communication is interrupted or the execution feedback timeout occurs; and a new scheduling instruction is received from the superior scheduling agency. When any of the preset triggering conditions is met continuously for at least one sampling period, or the cumulative number of times it is met reaches a preset number within a preset time window, the risk control agent generates a renegotiation optimization trigger message and sends it to the coordination agent. After receiving the renegotiation optimization trigger message, the coordination agent generates a differential scheduling task contract based on the non-execution time window, and re-negotiates the scheduling of each resource agent based on the differential scheduling task contract to obtain the re-scheduling negotiation result.

[0054] As an optional implementation, before the edge execution agent issues the control command to each edge execution device, it performs a local validity check on the control command. If the local validity check fails, the edge execution agent generates an abnormal receipt and triggers renegotiation or degraded control. The local validity check includes: power upper and lower limit check, device operating status check, energy storage charge status check, communication status check, safety interlock check, and user-side constraint check. The control command includes: resource device identifier, command type, target active power, target reactive power, start execution time, end execution time, ramp rate limit, maximum allowable deviation, command validity period, safety interlock conditions, and abnormal rollback strategy.

[0055] As an optional implementation, in step S6, the full-process traceable audit of scheduling includes: binding each scheduling task contract, capability quotation message, scheduling negotiation result, scheduling conflict arbitration priority score, control command, real-time device execution data, and rescheduling negotiation result with a unique identifier, timestamp, digital signature, and status version number, and generating audit index log records according to the scheduling task number; generating a corresponding hash digest for each audit index log record, and writing the hash digest corresponding to the previous audit index log record into the next audit index log record to form a chain audit record; the chain audit record is used to trace the source of scheduling decisions, optimization basis, command execution status, and anomaly handling results.

[0056] This application first addresses the heterogeneous resource access and state normalization of the virtual power plant: Data from distributed power sources, energy storage devices, charging facilities, adjustable loads, demand response terminals, distribution network nodes, and electricity market signals within the virtual power plant are collected to generate a unified resource state vector, which is then encapsulated into a Hermes-like scheduling message envelope. Next, multi-agent role construction and scheduling task contract generation are performed: based on the virtual power plant's resource types, scheduling objectives, and operational constraints, resource agents, predictive agents, market agents, grid security agents, carbon constraint agents, and coordination agents are constructed. The system comprises a risk control agent and an edge execution agent, which parse external scheduling requirements into scheduling task contracts that include power demand, time scale, constraints, response priorities, and settlement rules. Negotiation and constraint propagation are then performed based on a Hermes-like message bus: the scheduling task contracts are published via the Hermes-like message bus, and each agent generates capability quote messages, feasible domain boundaries, and risk warnings based on its own state, adjustable capabilities, and local constraints. Scheduling negotiation is completed through message subscription, version synchronization, conflict detection, and acknowledgment mechanisms. Next, multi-objective safety constraint scheduling optimization and conflict arbitration are performed: the coordinating agent summarizes the capability quote messages, feasible domain boundaries, and risk warnings returned by each agent, constructs a multi-objective safety constraint scheduling optimization model, and jointly optimizes operating costs, supply-demand deviations, grid security risks, carbon emissions, equipment lifespan losses, and market returns. Conflict arbitration is performed when multiple scheduling tasks or multiple agents request the same adjustment capability, generating a decomposed scheduling plan for the virtual power plant. Finally, edge-cloud collaborative instruction issuance and closed-loop self-correction are performed: the decomposed scheduling plan is converted into control instructions for each resource device, issued to the corresponding devices through the edge execution agent, and execution feedback is collected. Real-time telemetry data and deviation alarms; when execution deviation, prediction error, power grid alarm, market signal change, or communication anomaly meets preset trigger conditions, rolling re-optimization and event-driven multi-agent renegotiation are initiated; finally, the entire scheduling process is audited and traced: the scheduling task contract, version changes of the unified resource state vector, capacity quotation messages, constraint propagation messages, optimization solution results, conflict arbitration basis, control instructions, execution feedback, anomaly alarms, and re-optimization trigger events are recorded with timestamps in the audit index log, and a traceable link is established based on message identifier, state version number, and digital signature.

[0057] In an exemplary embodiment, to illustrate the technical solutions provided in the embodiments of this application in detail, the following are provided: Figure 3This paper presents a Hermes-based multi-agent scheduling method for virtual power plants, applied to a day-ahead planning and real-time rolling scheduling scenario for a virtual power plant in an industrial park. The virtual power plant's access resources include: 30MW of distributed photovoltaic capacity, 10MW of distributed wind power capacity, 12MW / 24MWh of electrochemical energy storage system, 6MW of adjustable power for charging pile clusters, 5MW of adjustable capacity for building air conditioning and temperature control loads, and 8MW of adjustable capacity for industrial interruptible loads. The park's distribution network includes a 35kV main transformer, 10kV feeders, multiple grid-connected nodes, and smart meter terminals. The virtual power plant simultaneously participates in the day-ahead energy market, real-time demand response, and peak shaving / valley filling operations. The method specifically includes the following implementation steps: Step 1: Access and normalization of heterogeneous resource data in virtual power plant: Collect data from distributed power sources, energy storage devices, charging facilities, adjustable loads, demand response terminals, distribution network nodes and electricity market signals within the virtual power plant, construct a unified resource state vector, and encapsulate the unified resource state vector into a Hermes-like scheduling message envelope.

[0058] In this embodiment of the application, the unified resource state vector can be represented as: .

[0059] in, Indicates the first One resource in The state vector at time t, For the first Each resource has a unique identifier. For the first The resource type corresponding to each resource For the first Each resource corresponds to a grid-connected node. and The first The maximum and minimum adjustable power corresponding to each resource. and The first The corresponding upward and downward ramp rates for each resource. For the first The sustainable response time corresponding to each resource For the first The state of charge of the energy storage resources corresponding to each resource. For the first The cost of calling each resource. For the first The predictive uncertainty corresponding to each resource For the first The carbon emission factor corresponding to each resource For the first The user comfort or production impact parameters corresponding to each resource For the first Equipment lifespan loss parameters corresponding to each resource. For the first Communication availability corresponding to each resource For the first The state confidence level corresponding to each resource. For the first Security constraint labels corresponding to each resource.

[0060] In this embodiment of the application, the Hermes-like scheduling message envelope can be represented as: .

[0061] in, Represents a Hermes-class scheduling message envelope. For message identification, To send the intelligent agent, In order to receive the intelligent agent, To subscribe to the topic, For message type, For message timestamps, For scheduling time windows, For message payload, For a constraint set, For confidence level, Priority This is the status version number. To confirm the receipt identifier, For digital signatures, This is a flag for exception handling.

[0062] like Figure 3 As shown, various resource status data are first collected through SCADA systems, energy management systems, charging operation platforms, building automation systems, energy storage converters, smart meters, and meteorological data interfaces. The collected data undergoes time synchronization, outlier removal, missing value completion, and unit normalization. For example, photovoltaic output, energy storage charging and discharging power, charging pile load, and interruptible industrial load capacity are uniformly converted into a power sequence at the same scheduling time granularity.

[0063] For each resource, a unified resource state vector is constructed. Taking energy storage resources as an example, its state vector includes energy storage number, grid-connected node, maximum charging power, maximum discharging power, current SOC, upper SOC limit, lower SOC limit, charging / discharging efficiency, cycle life loss coefficient, current availability, communication status, and response confidence. Taking charging pile clusters as an example, its state vector includes site number, number of connected vehicles, estimated departure time, user's expected charging capacity, current charging power, adjustable power, adjustable power, user satisfaction boundary, and communication availability. Subsequently, the resource state vector is encapsulated into a Hermes-like scheduling message envelope. For example, the energy storage agent publishes a state update message to the coordinating agent. The message contains the resource ID, message type, state version number, current SOC, adjustable power range, sustainable response duration, life loss cost, and digital signature. After receiving the state update message, the coordinating agent verifies the state version and updates the resource state cache.

[0064] Step 2: Multi-agent role construction and scheduling task contract generation: Based on the resource type, scheduling objectives and operational constraints of the virtual power plant, construct resource agents, prediction agents, market agents, grid security agents, carbon constraint agents, coordination agents, risk control agents and edge execution agents, and parse external scheduling requirements into scheduling task contracts.

[0065] In this embodiment of the application, the scheduling task contract includes task number, scheduling source, target power curve, response start time, response duration, minimum response capacity, maximum response capacity, reserve capacity requirement, voltage or power flow safety constraints, market settlement price, default penalty, carbon emission constraints, resource type preference, response confirmation deadline and abnormal degradation strategy.

[0066] like Figure 4 As shown, the embodiments of this application construct the following intelligent agents for a virtual power plant: (1) Resource agents correspond to resources such as photovoltaic, wind power, energy storage, charging piles, air conditioning loads, and industrial interruptible loads, respectively. They are responsible for maintaining the local resource status, calculating adjustable capacity, generating capacity quotations, and executing scheduling commitments.

[0067] (2) Based on weather forecasts, historical power output curves, historical load curves and user behavior data, the predictive agent generates new energy power output forecasts, load forecasts and charging demand forecasts for the next 24 hours, and outputs the prediction confidence interval.

[0068] (3) The market agent analyzes the day-ahead market price, real-time electricity price, demand response subsidy, ancillary service price and default penalty rules to generate the market revenue boundary.

[0069] (4) The power grid security intelligent agent generates scheduling security constraints based on the distribution network topology, line capacity, node voltage, main transformer load rate and power flow calculation results.

[0070] (5) The carbon-constrained agent calculates the carbon emission intensity corresponding to different resource call combinations based on resource type and external carbon emission factor.

[0071] (6) The coordinating agent is responsible for task contract release, capability quotation summary, multi-objective optimization solution, conflict arbitration and scheduling plan generation.

[0072] (7) The risk control agent is responsible for detecting risks such as communication interruption, execution timeout, resource default, abnormal pricing and scheduling infeasibility.

[0073] (8) The edge execution agent is deployed on the edge gateway or resource controller side of the park and is responsible for converting the scheduling plan into device control instructions and collecting execution feedback.

[0074] When a virtual power plant receives a demand response task stating "reduce the grid-connected power of the industrial park by 10MW between 18:00 and 20:00, with a response duration of 2 hours," the coordinating agent parses this external demand into a scheduling task contract. This contract includes the task number, response time window, target power reduction, minimum acceptable response capacity, response confirmation deadline, market subsidy price, penalty for breach of contract, grid security constraints, and anomaly degradation strategy, and is published to relevant resource agents and constraint agents via a Hermes-like message bus.

[0075] Step 3: Negotiation and Constraint Propagation Based on Hermes-like Message Bus: The scheduling task contract is published through the Hermes-like message bus. Each agent generates a capability quote message, feasible domain boundary and risk warning based on its own state, adjustable capabilities and local constraints, and completes the scheduling negotiation through message subscription, version synchronization, conflict detection and acknowledgment mechanism.

[0076] In this embodiment, after receiving the scheduling task contract, each resource agent first determines whether it has the capability to respond based on its own resource state vector. If it has the capability to respond, it returns a capability quote message, which includes the responsive power range, response cost, response confidence, expected execution deviation, equipment lifespan loss, impact on user comfort, and local security constraints. If it does not have the capability to respond, it returns a rejection message and the reason for rejection.

[0077] In this embodiment, after receiving a capability quote message, the coordinating agent compares the status version number in the capability quote message with the latest status version number in the resource status cache. If the versions match, the capability quote message is included in the candidate resource combination; if the versions do not match, the capability quote message is discarded, and a requote request is sent to the corresponding resource agent.

[0078] In this embodiment, after receiving the scheduling task contract, the resource agent calculates its own capability to participate in the response. Based on the current SOC of 65%, the lower limit of SOC of 20%, the maximum discharge power of 12MW, and the current operating temperature being normal, the energy storage agent returns a capability quotation message stating that it can provide up to 8MW of discharge support between 18:00 and 20:00, with a response confidence level of 95% and a lifetime loss cost of 0.08 yuan / kWh.

[0079] The charging pile intelligent agent returns a capability quote message based on the vehicle's departure time and the user's expected charge level: "3MW charging load can be reduced for 2 hours, with a moderate impact on user satisfaction and a response confidence level of 82%." The air conditioning load intelligent agent returns a capability quote message: "2MW can be reduced within comfort constraints for 1.5 hours, with room temperature deviation not exceeding 1.5℃." The industrial interruptible load intelligent agent returns a capability quote message: "5MW can be interrupted for 2 hours, but requires 20 minutes' advance confirmation, resulting in higher response costs." The grid safety intelligent agent, based on distribution network power flow calculations, detects a potential overload risk on a 10kV feeder around 18:00 and sends a constraint propagation message to the coordinating intelligent agent, requiring that the reverse power flow of downstream resources of this feeder not exceed a preset limit. The market intelligent agent simultaneously releases demand response subsidy prices and default penalty information. The risk control intelligent agent detects a decline in communication quality at some charging stations and lowers their response confidence level. The coordinating intelligent agent completes the subscription, aggregation, version verification, and confirmation receipt of the above messages through a Hermes-like message bus. When a resource quote message is inconsistent with its latest state version, the coordinating agent sends a requote request to the corresponding resource agent to avoid using the expired state to generate a scheduling plan.

[0080] Step 4: Multi-objective safety constraint scheduling optimization and conflict arbitration: The coordinating agent summarizes the capability quotation messages, feasible domain boundaries and risk warnings, constructs a multi-objective safety constraint scheduling optimization model, and generates a virtual power plant decomposition scheduling plan.

[0081] In this embodiment of the application, the objective function of the multi-objective security constraint scheduling optimization model is: .

[0082] in, To achieve the overall scheduling target value, For resource allocation and operation and maintenance costs, For the target power deviation, For power grid security risks, For carbon emission costs, This is to account for the lifespan loss of energy storage and controllable equipment. For market profits, , , , , and These are weighted coefficients corresponding to resource allocation and operation and maintenance costs, target power deviation, grid security risks, carbon emission costs, energy storage and controllable equipment lifespan losses, and market revenue. By dynamically adjusting these weighted coefficients, different scheduling objectives can be adapted to, such as peak shaving and valley filling, renewable energy consumption, ancillary services, demand response, carbon emission reduction, and emergency supply.

[0083] In this embodiment, the power flow constraints of the distribution network are calculated based on the distribution network topology, node injected power, line impedance, transformer capacity, and switch status; the node voltage constraints include the voltage amplitude of each node being within a preset upper and lower limit range; and the line capacity constraints include the line current or line load rate not exceeding the corresponding line rated capacity.

[0084] In this embodiment of the application, the scheduling conflict arbitration priority score can be expressed as: .

[0085] in, Scoring is done based on scheduling priority. For the sake of power grid safety and urgency, For market default losses, For resource response reliability, Due to the urgency of scheduling, Contribution to carbon constraints For the sake of user comfort, , , , , , These are weighted coefficients corresponding to the urgency of grid security, market default losses, resource response reliability, scheduling time urgency, carbon constraint contribution, and user comfort impact. A higher priority score indicates a higher execution priority for the corresponding task in resource conflicts. When multiple agents simultaneously request the same resource, or when the adjustment capacity of the same resource is insufficient to satisfy multiple scheduling tasks, the scheduling task with the higher priority score is prioritized, and the remaining scheduling tasks undergo alternative resource search, power reduction, time migration, or manual review.

[0086] like Figure 5 As shown, after the coordinating agent aggregates the information returned by each agent, it constructs a candidate resource pool and establishes a multi-objective safety constraint scheduling optimization model. The optimization objectives of this application embodiment include: minimizing resource call costs, minimizing target power reduction deviations, minimizing feeder overload and voltage over-limit risks, minimizing energy storage lifetime losses, maximizing demand response subsidy benefits, and reducing the impact on user comfort.

[0087] During the solution process, the model needs to meet the following constraints: the total power reduction of the virtual power plant between 18:00 and 20:00 shall not be less than 10MW; the energy storage SOC shall be between 20% and 90%; the energy storage discharge power shall not exceed 12MW; the power reduction of the charging pile shall not cause the user's expected off-site power to be lower than the promised value; the air conditioning load adjustment shall not cause the room temperature to exceed the comfort range; the number of interruptible industrial load calls shall not exceed the agreed number for the day; and the feeder current and node voltage shall meet the distribution network safety boundary.

[0088] If an energy storage agent is simultaneously invoked by both demand response and reserve capacity tasks, the coordinating agent initiates conflict arbitration. Priority scores are calculated based on grid security urgency, market default losses, resource response reliability, dispatch time urgency, carbon constraint contribution, and user comfort impact. If the grid security urgency of the reserve capacity task is higher than that of the demand response task, a portion of the energy storage capacity is reserved for backup services, and the demand response task shortfall is transferred to industrial interruptible loads and air conditioning loads to meet overall dispatch objectives.

[0089] The final virtual power plant decomposed dispatch plan is as follows: the energy storage system will discharge 6MW from 18:00 to 20:00; the charging pile cluster will reduce 2MW; the air conditioning load will be reduced by 1MW; and the industrial interruptible load will be interrupted by 1MW, for a total power reduction of 10MW. At the same time, 2MW of energy storage reserve capacity will be reserved for emergency support of feeder heavy loads.

[0090] Step 5: Edge Collaborative Command Issuance and Closed-Loop Self-Correction: The virtual power plant's decomposed scheduling plan is converted into control commands for each resource device, which are then issued to the corresponding devices through edge execution agents. Execution feedback, real-time telemetry data, and deviation alarms are collected. When preset trigger conditions are met, rolling re-optimization and event-driven multi-agent renegotiation are initiated.

[0091] In this embodiment, closed-loop self-calibration includes: the edge execution agent collecting actual power, device status, and execution confirmation information of the device according to a preset sampling period; the coordination agent calculating the deviation between the actual power curve and the planned power curve; when the deviation exceeds a threshold, generating a differential scheduling task contract, only performing rolling optimization on the remaining scheduling targets within the unexecuted time window, without recalculating the time period that has already been executed.

[0092] In this embodiment, the preset deviation threshold can be determined based on a fixed proportion of the target power of the scheduling task, the absolute power value, or a combination of both; the prediction error threshold can be determined based on the confidence interval of new energy prediction; the price threshold can be determined based on the rate of change of real-time market price within two consecutive scheduling cycles; and the execution feedback timeout threshold can be determined based on the resource communication link type and scheduling time granularity.

[0093] In this embodiment, the edge execution agent receives and sends control commands through one or more communication protocols selected from IEC104, Modbus, MQTT, OPCUA, DL / T645, or the charging pile operation platform interface.

[0094] In this embodiment, the coordinating agent converts the decomposed scheduling plan into device-level control commands, which are then distributed to the energy storage PCS, charging station controller, building air conditioning control system, and industrial load control terminal via edge execution agents. The edge execution agents locally verify the validity of the commands, including power upper and lower limits, device operating status, communication status, and safety interlock conditions. After successful verification, the edge execution agents execute the control commands and periodically transmit back the executed power, device status, execution confirmation, and anomaly alarms.

[0095] At 18:35, the predictive agent detected that the photovoltaic output decreased by more than the prediction error threshold due to cloud cover, and the actual grid-connected power deviation reached the preset deviation threshold. The risk control agent immediately issued a re-optimization trigger message, and the coordinating agent initiated event-driven multi-agent renegotiation. The energy storage agent recalculated the discharge capacity, the charging pile agent updated vehicle departure constraints, the air conditioning load agent updated room temperature status, and the grid security agent updated feeder power flow constraints. Based on the new state version, the coordinating agent performed rolling optimization, adjusting the energy storage discharge power from 6MW to 7MW and the air conditioning load reduction power from 1MW to 1.5MW to compensate for the power deviation caused by the decrease in renewable energy output.

[0096] Step 6: Full-Process Audit and Traceability of Scheduling: Record the scheduling task contract, agent bidding, constraint changes, optimization results, conflict arbitration reasons, control instructions, execution feedback, and anomaly handling process. Each scheduling task generates a unique task number, each message generates a message identifier, changes in the same resource state generate an incrementing state version number, and each control instruction records the instruction issuance time, execution confirmation time, feedback value, and anomaly reason. When conflict arbitration occurs, record the task number, resource number, priority score, scoring factor, and final arbitration result of the involved tasks.

[0097] In this embodiment, a hash digest is generated for each message and each scheduling decision, and the hash digest of the previous audit index log record is written into the next audit index log record to form a chain of audit records; at the same time, the consistency of scheduling tasks, capacity quotations, optimization results, instruction issuance and execution feedback is verified by combining the digital signature, timestamp, status version number and confirmation receipt identifier in the message envelope.

[0098] Figure 6This paper presents an architecture for a multi-agent scheduling system for a virtual power plant based on a Hermes-like framework. The architecture consists of four layers from top to bottom, with a Hermes-like message interface running through all layers to enable cross-layer message interaction. The first layer is the data and resource access layer, connecting distributed power sources, energy storage, charging piles, adjustable loads, metering, and communication terminals. The second layer is the agent collaboration layer, deploying resource agents, prediction agents, market agents, safety agents, and carbon constraint agents to achieve multi-agent collaborative interaction. The third layer is the optimization and risk control layer, implementing multi-objective optimization, task contract management, conflict arbitration, limit violation verification, and policy rollback. The fourth layer is the execution and audit layer, responsible for edge control, receipt confirmation, anomaly degradation handling, and generating chained audit records. Each message is bound to a message identifier, timestamp, status version number, and digital signature by recording task contracts, agent quotations, constraint changes, optimization results, conflict arbitration reasons, control instructions, execution feedback, and anomaly handling processes. Each scheduling task generates an independent audit index, enabling operations personnel to trace the source, basis, and execution results of each scheduling decision.

[0099] In this embodiment, if a resource conflict arbitration occurs, the system records the task number, resource number, each scoring factor, final priority score, and arbitration result of the conflicting task; if an edge execution failure occurs, the system records the control instruction content, verification failure reason, exception receipt, and renegotiation result. Audit logs are stored using hash digests and chained recording methods, and can be used for post-event verification, market settlement dispute handling, and scheduling security responsibility tracing.

[0100] In this embodiment of the application, during the simulation scheduling test of the virtual power plant in the park, compared with the traditional centralized single optimization scheme, the scheduling plan generation time can be significantly shortened, the number of unexecutable scheduling due to resource status expiration is reduced, the actual response power deviation is reduced, the renewable energy absorption capacity is improved, the number of ineffective energy storage cycles is reduced, and scheduling messages and control commands can be traced through message identifiers, status version numbers and digital signatures, which can meet the real-time scheduling and market-oriented operation needs of the park-level virtual power plant.

[0101] The multi-agent scheduling method for virtual power plants based on Hermes-like architecture proposed in this application has the following technical advantages: (1) By using a unified resource state vector and Hermes-like scheduling message envelope, heterogeneous objects such as sources, grids, loads, storage, and charging in the virtual power plant are modeled in a unified manner and their communication is standardized. This solves the problems of inconsistent resource state semantics, scattered scheduling interfaces, and difficulties in cross-device collaboration in traditional virtual power plant systems, thereby improving resource access efficiency and scheduling data consistency.

[0102] (2) By introducing resource agents, predictive agents, market agents, power grid security agents, carbon constraint agents, coordination agents, risk control agents and edge execution agents, the scheduling task can be transformed from centralized single-point calculation to multi-agent negotiation and distributed response. This can reduce the computing pressure on the central scheduling platform, shorten the real-time scheduling response delay, and adapt to large-scale distributed resource high-frequency scheduling scenarios.

[0103] (3) By scheduling task contracts, capability pricing, constraint propagation, confirmation receipts and state version synchronization mechanisms, each agent can clearly express adjustable capabilities, local constraints, response costs and risk information during the scheduling process, avoiding the problems of unclear negotiation semantics, delayed conflict discovery and untraceable execution in traditional multi-agent schemes.

[0104] (4) By constructing a multi-objective safety constraint scheduling optimization model, operating costs, supply and demand deviation, grid security risks, carbon emissions, equipment life loss and market revenue are jointly optimized, and constraints such as distribution network power flow, voltage, line capacity, energy storage SOC and user comfort are explicitly introduced, so that the virtual power plant scheduling results can meet both the market revenue target and the grid security and user-side executability requirements.

[0105] (5) Through edge-cloud collaborative instruction issuance and closed-loop self-correction mechanism, when actual execution deviation, prediction error, grid alarm, market price change or resource communication anomaly occurs, rolling re-optimization and event-driven multi-agent renegotiation can be automatically triggered to avoid full manual intervention and improve the robustness and adaptability of virtual power plant in complex operating environment.

[0106] (6) Through digital signatures, status version numbers, confirmation receipts, hash digests and chain audit records, the entire process of scheduling tasks, capacity quotation, constraint propagation, optimization solution, conflict arbitration, instruction issuance and execution feedback is recorded, so that the source of virtual power plant scheduling decisions, optimization basis, execution status and abnormal handling results are traceable.

[0107] (7) Combining artificial intelligence multi-agent algorithms, message bus communication mechanisms, distribution network physical constraints, edge control execution and scheduling auditing technologies can directly improve the computational efficiency, communication efficiency, control real-time performance and operational security of the virtual power plant scheduling system, and has strong engineering applicability.

[0108] Based on the same inventive concept, this application also provides a Hermes-based virtual power plant multi-agent scheduling device for implementing the Hermes-based virtual power plant multi-agent scheduling method described above. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations in the embodiments of the Hermes-based virtual power plant multi-agent scheduling device provided below can be found in the limitations of the Hermes-based virtual power plant multi-agent scheduling method described above, and will not be repeated here.

[0109] In one exemplary embodiment, such as Figure 7 As shown, a Hermes-based virtual power plant multi-agent scheduling device is provided, specifically including: The data acquisition module is used to acquire heterogeneous resource data of the virtual power plant.

[0110] The data encapsulation module is used to construct a unified resource state vector based on the heterogeneous resource data of the virtual power plant and encapsulate it into a Hermes-like scheduling message envelope.

[0111] The scheduling task contract parsing, distribution, and scheduling negotiation module is used to construct a multi-agent scheduling cluster based on the Hermes-like scheduling message envelope, parse external scheduling requirements into scheduling task contracts, and distribute them to each agent in the multi-agent scheduling cluster. This enables each agent to generate capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustability, and local constraints, and to complete scheduling negotiation based on the messenger-style multi-agent collaborative message mechanism to obtain the scheduling negotiation result.

[0112] The conflict arbitration and plan generation module is used to construct a multi-objective safety constraint scheduling optimization model based on the capability quotation message, feasible domain boundary and risk warning, and to perform scheduling conflict arbitration priority scoring to generate a virtual power plant decomposition scheduling plan.

[0113] The instruction generation, distribution, and renegotiation optimization module is used to generate control instructions based on the virtual power plant's decomposed scheduling plan, and distribute them to each edge execution device through the edge execution agents in the multi-agent scheduling cluster. It collects real-time execution data of each edge execution device and automatically starts local rolling renegotiation optimization when preset trigger conditions are met to obtain the rescheduling negotiation result.

[0114] The scheduling process traceability and auditing module is used to attach unique identifiers, timestamps, digital signatures, and status version numbers to the scheduling task contract, the capability quotation message, the scheduling negotiation result, the scheduling conflict arbitration priority score, the control command, the device real-time execution data, and the rescheduling negotiation result, and to achieve full-process traceability and auditing of the scheduling process using a chain hash notarization method.

[0115] This application proposes a Hermes-like multi-agent scheduling method and device for virtual power plants. Addressing the challenges of diverse resources (distributed photovoltaic, wind power, energy storage, charging facilities, adjustable loads, and demand response) in virtual power plants, including rapid state changes, complex scheduling constraints, high latency in centralized scheduling, low cross-resource collaboration efficiency, and difficulty in auditing the scheduling process, a Hermes-like messenger-style multi-agent collaborative messaging mechanism is designed. This mechanism standardizes the expression of various resources, market signals, and distribution network security constraints within the virtual power plant through unified resource state vectors, scheduling task contracts, and scheduling message envelopes. Through the collaborative work of resource agents, prediction agents, market agents, grid security agents, carbon constraint agents, coordination agents, risk control agents, and edge execution agents, scheduling task issuance, capacity quotation, constraint propagation, version synchronization, conflict arbitration, edge collaborative execution, and closed-loop self-correction are achieved. Furthermore, a multi-objective security constraint scheduling optimization model is used to jointly optimize operating costs, supply-demand deviations, grid security risks, carbon emissions, equipment lifespan depletion, and market returns. The embodiments of this application can improve the real-time dispatch response speed, heterogeneous resource aggregation efficiency, new energy consumption capacity and dispatch process traceability of virtual power plants, and are applicable to day-ahead planning, real-time dispatch, ancillary services, peak shaving and valley filling and demand response scenarios of virtual power plants at the park level, regional level and city level.

[0116] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 8As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores heterogeneous resource data for a virtual power plant. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a Hermes-based multi-agent scheduling method for a virtual power plant.

[0117] Figure 8 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0118] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0119] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0120] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations and be authorized by the owner of the corresponding device.

[0121] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0122] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0123] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0124] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A multi-agent scheduling method for a virtual power plant based on a Hermes-like framework, characterized in that, The Hermes-like mechanism is a messenger-based multi-agent collaborative messaging mechanism for energy scheduling scenarios. It possesses end-to-end capabilities including unified message envelope encapsulation, standardized scheduling contract distribution, multi-agent distributed negotiation, resource status version synchronization, message receipt confirmation, and closed-loop handling of abnormal events. The method includes: Acquire heterogeneous resource data of virtual power plants; Based on the heterogeneous resource data of the virtual power plant, a unified resource state vector is constructed and encapsulated as a Hermes-like scheduling message envelope; A multi-agent scheduling cluster is constructed based on the Hermes-like scheduling message envelope. External scheduling requirements are parsed into scheduling task contracts and distributed to each agent in the multi-agent scheduling cluster. Each agent generates capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustability, and local constraints. Scheduling negotiation is completed based on the messenger-style multi-agent collaborative message mechanism to obtain the scheduling negotiation result. Based on the aforementioned capability pricing message, feasible domain boundary, and risk warning, a multi-objective security constraint scheduling optimization model is constructed, and scheduling conflict arbitration priority scoring is performed to generate a virtual power plant decompositional scheduling plan. Control commands are generated based on the decomposed scheduling plan of the virtual power plant and distributed to each edge execution device through the edge execution agent in the multi-agent scheduling cluster. Real-time execution data of each edge execution device is collected. When the preset triggering conditions are met, local rolling renegotiation optimization is automatically started to obtain the rescheduling negotiation result. Each of the scheduling task contract, the capability quotation message, the scheduling negotiation result, the scheduling conflict arbitration priority score, the control command, the device real-time execution data, and the rescheduling negotiation result is assigned a unique identifier, a timestamp, a digital signature, and a status version number. A chain hash notarization method is used to achieve full-process traceability and auditing of the scheduling process.

2. The multi-agent scheduling method for virtual power plants based on Hermes-like architecture according to claim 1, characterized in that, A multi-agent scheduling cluster is constructed based on the Hermes-like scheduling message envelope. External scheduling requirements are parsed into scheduling task contracts and distributed to each agent in the multi-agent scheduling cluster. This allows each agent to generate capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustability, and local constraints. Scheduling negotiation is then completed based on the messenger-style multi-agent cooperative messaging mechanism to obtain the scheduling negotiation result, specifically including: Based on the Hermes-like scheduling message envelope, and combined with the resource type, scheduling objective, and operational constraints of the virtual power plant, a multi-agent scheduling cluster is constructed. Based on the multi-agent scheduling cluster, external scheduling requests are received from the power grid, the electricity market, and the demand response side. The external scheduling requirements are parsed to obtain a scheduling task contract; the scheduling task contract includes: power requirements, time scale, constraints, response priority, and settlement rules; By utilizing the standardized scheduling contract distribution capability of the messenger-based multi-agent collaborative messaging mechanism, the scheduling task contract is distributed to each agent in the multi-agent scheduling cluster, so that each agent can generate corresponding capability quote messages, feasible domain boundaries, and risk warnings based on its own resource status, adjustable capabilities, and local constraints. Based on the capability quote messages, feasible domain boundaries, and risk warnings corresponding to each agent, the scheduling negotiation between the agents is carried out using the multi-agent distributed negotiation capabilities, resource state version synchronization capabilities, message receipt confirmation capabilities, and abnormal event closed-loop processing capabilities of the messenger-style multi-agent collaborative messaging mechanism, so as to obtain the scheduling negotiation results.

3. The multi-agent scheduling method for virtual power plants based on Hermes-like architecture according to claim 1 or 2, characterized in that, The multi-agent scheduling cluster includes: resource agents, prediction agents, market agents, power grid security agents, carbon constraint agents, coordination agents, risk control agents, and edge execution agents; The resource agent is used to maintain the unified resource state vector of the heterogeneous resources of the corresponding virtual power plant, and to generate the capability quotation message, feasible domain boundary and risk warning based on the scheduling task contract, its own resource state, adjustable capabilities and local constraints. The predictive agent is used to generate new energy output prediction results, load prediction results, charging demand prediction results, and response availability prediction results based on meteorological data, historical output data, historical load data, and user behavior data. The market intelligence agent is used to analyze day-ahead market, real-time market, ancillary services market, and demand response subsidy signals; The power grid security intelligent agent is used to generate power flow constraints, voltage over-limit constraints, line capacity constraints, and reserve capacity constraints for the distribution network. The carbon-constrained agent is used to calculate the carbon emission intensity and carbon constraint boundary corresponding to resource allocation; The coordinating agent is used to parse the external scheduling requirements to obtain the scheduling task contract and distribute it to each of the resource agents. It also summarizes the capability quotation messages, feasible domain boundaries and risk warnings fed back by each of the resource agents and verifies the status version number. The agent constructs the multi-objective security constraint scheduling optimization model and scores the scheduling conflict arbitration priority of each of the resource agents to generate a virtual power plant decomposition scheduling plan. The risk control intelligent agent is used to identify risks such as communication anomalies, equipment anomalies, response defaults, and scheduling infeasibility. The edge execution agent is used for controlling the validity verification and issuance of control commands, controlling edge execution devices, and collecting real-time execution data from the devices.

4. The multi-agent scheduling method for virtual power plants based on Hermes-like architecture according to claim 3, characterized in that, After receiving the capability quote message from the resource agent, the coordinating agent compares the status version number in the capability quote message with the latest status version number in the resource status cache; if the status version number matches the latest status version number, the capability quote message is included in the candidate resource combination. Otherwise, the capability quote message is discarded, and a requote request is sent to the corresponding resource agent to avoid generating an unexecutable virtual power plant decomposition scheduling plan using expired states.

5. The multi-agent scheduling method for virtual power plants based on Hermes-like architecture according to claim 3, characterized in that, The preset triggering conditions include at least one of the following: the deviation between the actual executed power and the planned power is greater than a preset deviation threshold; the prediction error of new energy output is greater than a preset prediction error threshold; the load prediction error is greater than a preset load error threshold; the energy storage state of charge is less than or greater than a preset safety boundary; the voltage of the distribution network node or the line current triggers an over-limit alarm; the real-time market price fluctuation is greater than a preset price threshold; the resource intelligent agent communication is interrupted or the execution feedback timeout occurs; or a new dispatching instruction is received from the superior dispatching agency. When any of the preset trigger conditions is met continuously for at least one sampling period, or when the number of times the conditions are met reaches a preset number within a preset time window, the risk control agent generates a renegotiation optimization trigger message and sends it to the coordination agent. After receiving the renegotiation optimization trigger message, the coordination agent generates a differential scheduling task contract based on the non-execution time window, and re-negotiates the scheduling of each resource agent based on the differential scheduling task contract to obtain the re-scheduling negotiation result.

6. The multi-agent scheduling method for virtual power plants based on Hermes-like architecture according to claim 1, characterized in that, The objective function of the multi-objective security constraint scheduling optimization model is: ； in, To achieve the overall scheduling target value, For resource allocation and operation and maintenance costs, For the target power deviation, For power grid security risks, For carbon emission costs, This is to account for the lifespan loss of energy storage and controllable equipment. For market profits, , , , , and These are the weighting coefficients corresponding to resource allocation and operation and maintenance costs, target power deviation, grid security risks, carbon emission costs, energy storage and controllable equipment lifespan losses, and market revenue.

7. The multi-agent scheduling method for virtual power plants based on Hermes-like architecture according to claim 3, characterized in that, Before issuing the control commands to each edge execution device, the edge execution agent performs local validity checks on the control commands. If the local validity check fails, the edge execution agent generates an exception receipt and triggers renegotiation or downgrade control. The local validity checks include: power upper and lower limit checks, device operating status checks, energy storage charge status checks, communication status checks, safety interlock checks, and user-side constraint checks. The control commands include: resource device identifier, command type, target active power, target reactive power, start execution time, end execution time, ramp rate limit, maximum allowable deviation, command validity period, safety interlock conditions, and exception fallback strategy. The full-process traceable audit of scheduling includes: binding each scheduling task contract, capability quotation message, scheduling negotiation result, scheduling conflict arbitration priority score, control command, real-time device execution data, and rescheduling negotiation result with a unique identifier, timestamp, digital signature, and status version number, and generating audit index log records according to the scheduling task number; generating a corresponding hash digest for each audit index log record, and writing the hash digest corresponding to the previous audit index log record into the next audit index log record to form a chain of audit records; the chain of audit records is used to trace the source of scheduling decisions, optimization basis, command execution status, and anomaly handling results.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the Hermes-based virtual power plant multi-agent scheduling method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the Hermes-based virtual power plant multi-agent scheduling method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the Hermes-based virtual power plant multi-agent scheduling method according to any one of claims 1-7.

Images