A large model multi-agent a2a virtual power grid dispatching power selling system and method

Abstract

The application provides a large model multi-agent A2A virtual power grid dispatching and power selling system and method, which comprises a vertical field energy large model module, an MCP tool layer module, an A2A multi-agent negotiation layer module and an application service layer module; the vertical field energy large model module serves as a system core reasoning engine; the large model multi-agent A2A virtual power grid dispatching and power selling system and method; the large model multi-agent A2A virtual power grid dispatching and power selling system provided by the application improves system operation safety and reliability, constructs a continuous learning closed loop of execution, evaluation, optimization and updating, iteratively optimizes dispatching and power selling strategies through review and optimization of agents, strengthens system adaptive and autonomous evolution capabilities, guarantees long-term stable and efficient operation, and provides a new technical path for intelligent operation of a virtual power grid.

Description

Technical Field

[0001] This invention relates to the field of intelligent dispatching of virtual power plants and electricity market transactions, and in particular to a large-scale multi-agent A2A virtual power grid dispatching and sales system and method. Background Technology

[0002] Virtual power plants integrate resources such as photovoltaic, wind power, energy storage, and flexible loads to participate in grid dispatch and electricity market transactions. The existing dispatch and electricity sales systems are disconnected. Dispatch relies on SCADA / EMS systems to ensure safety, while electricity sales rely on market platforms or manual operations to pursue profits. The two systems have data that cannot be communicated and decision-making is disconnected. Traditional dispatch often uses centralized mathematical models such as game theory and linear programming, which rely on manual construction of objective functions. In the face of dynamic scenarios such as fluctuations in new energy sources and sudden changes in electricity prices, the adaptive capability is insufficient.

[0003] Some solutions introduce AI-assisted scheduling, using large models to predict load and electricity prices, and then combining reinforcement learning to generate strategies. However, the large models only serve as prediction tools, and decision-making is still dominated by traditional models. This is a serial architecture and is not deeply integrated into the core of decision-making. Existing multi-agent systems use custom communication interfaces, making it difficult for agents from different vendors to interoperate. Expanding resources or entering the market requires a lot of customized code, resulting in poor openness and scalability.

[0004] Existing technologies suffer from four major pain points: the separation of scheduling and trading, non-standard communication between intelligent agents, shallow application of large models, and insufficient adaptation to complex environments. These issues make it difficult to coordinate heterogeneous resources, integrate scheduling and trading, limit the absorption rate of new energy and the revenue from electricity sales, and the A2A standardized protocol has not yet been applied to the field of virtual power grids. Furthermore, the lack of semantic-level communication and distributed negotiation mechanisms makes it impossible to support the understanding of unstructured information, online adaptation, and continuous learning.

[0005] Therefore, it is necessary to provide a large-scale multi-agent A2A virtual power grid dispatching and sales system and method to solve the above-mentioned technical problems. Summary of the Invention

[0006] This invention provides a large-scale multi-agent A2A virtual power grid dispatching and sales system and method, which solves four major pain points of the existing technology, namely, difficulty in resource coordination, difficulty in integrating dispatching and trading, lack of application of A2A protocol and lack of semantic communication and negotiation mechanism.

[0007] To address the aforementioned technical problems, the large-scale multi-agent A2A virtual power grid dispatching and sales system and method provided by this invention includes: a vertical domain energy large-scale model module, an MCP tool layer module, an A2A multi-agent negotiation layer module, and an application service layer module;

[0008] The vertical domain energy big model module, as the core reasoning engine of the system, relies on the power domain corpus for fine-tuning and has scheduling reasoning and market analysis capabilities. It provides semantic understanding, logical decision-making and unstructured information processing capabilities for each intelligent agent, replacing the traditional mathematical model and supporting adaptive and interpretable scheduling and endogenous power sales decisions.

[0009] The MCP tool layer module, as a standardized tool support layer, uses the MCP protocol to uniformly encapsulate tools such as power flow calculation, load price prediction, and energy storage estimation, and provides a unified calling interface. It does not require customized adaptation code and enables the tools to be used plug and play, providing accurate data and computational support for intelligent agent decision-making.

[0010] The A2A multi-agent negotiation layer module serves as the core of system decision-making. It deploys multiple types of agents and uses the A2A protocol to achieve registration discovery, task delegation, and multi-round semantic negotiation. It coordinates resource adjustability, electricity sales constraints, and security risks to reach consensus on scheduling and electricity sales, supporting heterogeneous resource collaboration and integrated scheduling and trading.

[0011] The application service layer module serves as the external service output layer. It integrates the core decision-making results of the system and outputs standardized services such as day-ahead scheduling, real-time adjustment, electricity sales pricing, and retrospective analysis. It supports visual display and interface with external systems, enabling the implementation of scheduling transaction results and operational decision support.

[0012] Preferably, the vertical domain energy big model module includes a base big model unit, an electric domain corpus unit, a continuous pre-training unit, a supervised fine-tuning unit, a semantic reasoning unit, and a decision generation unit.

[0013] Preferably, the MCP tool layer module includes an MCP protocol encapsulation unit, a power flow calculation solver unit, an electricity price prediction model unit, a load prediction model unit, an energy storage SOC estimator unit, a market information API unit, a meteorological interface unit, a constraint verification unit, and a unified call interface unit.

[0014] Preferably, the A2A multi-agent negotiation layer module includes a DER agent unit, a scheduling negotiation agent unit, a power sales strategy agent unit, a risk assessment agent unit, a user preference agent unit, a review and optimization agent unit, and an A2A protocol communication unit.

[0015] Preferably, the application service layer module includes a day-ahead scheduling planning unit, a real-time scheduling adjustment unit, an electricity sales price decision unit, a transaction review and analysis unit, a user energy consumption report unit, a visualization display unit, and a standardized interface unit;

[0016] The day-ahead dispatching planning unit outputs a 96-point day-ahead dispatching plan, clearly defining resource output, energy storage charging and discharging, and flexible load response instructions for each time period. This provides a standardized execution basis for the day-ahead operation of the virtual power plant, ensuring the orderly conduct of day-ahead dispatching. The real-time dispatching adjustment unit generates minute-level dispatching adjustment instructions in real time to address abnormal scenarios such as fluctuations in renewable energy output, sudden load changes, equipment failures, and sudden market price changes, ensuring real-time grid security and stable revenue. The electricity sales pricing decision-making unit outputs day-ahead, intraday, and spot market electricity sales pricing schemes, transaction volume allocation plans, and risk control measures, providing standardized decision-making support for electricity sales transactions and improving transaction efficiency and revenue. The transaction review and analysis unit generates a full-link review report for dispatching and transactions. Includes execution results, revenue analysis, deviation attribution, and optimization suggestions, providing data support for operational decisions and facilitating continuous improvement. The user energy consumption report unit generates reports on user energy consumption analysis, cost accounting, green electricity ratio, and optimization suggestions, tailored to user needs, enhancing user service experience and increasing user stickiness. The visualization unit provides a visual display interface for system architecture, resource status, scheduling schemes, transaction data, negotiation processes, risk information, and review results, intuitively presenting the system's operational status and facilitating operation and maintenance and decision-making. The standardized interface unit provides standardized external interfaces, supporting seamless integration with power grid dispatching systems, power market platforms, user-side management systems, and operation and maintenance platforms, improving system compatibility and integrability.

[0017] Preferably, the scheduling negotiation Agent unit includes a scheduling request receiving unit, a market signal receiving unit, an adjustable capability collection unit, an initial scheme generation unit, a multi-round semantic negotiation unit, a consensus decision-making unit, an A2A message routing unit, and a risk interaction unit. The A2A protocol communication unit further includes an Agent registration unit, an AgentCard management unit, a capability publishing unit, a standardization discovery unit, a task delegation unit, a semantic negotiation unit, a multi-round conversation unit, a conflict resolution unit, a message encoding / decoding unit, and a result feedback unit.

[0018] Preferably, the vertical domain energy big model module further includes a dynamic scenario adaptive reasoning unit, a multi-agent game equilibrium reasoning unit, and a long-cycle trend inference unit.

[0019] Preferably, the MCP tool layer module further includes a dynamic parameter self-calibration unit, a distributed resource health assessment unit, a multi-objective optimization solution unit, and a data security encrypted transmission unit;

[0020] Dynamic parameter self-calibration unit: Real-time calibration of load, electricity price, and new energy output prediction model parameters, correcting prediction deviations and improving data accuracy in dynamic scenarios; Distributed resource health assessment unit: Real-time monitoring of equipment health status, wear level, and aging trend, quantifying resource availability, and providing a basis for scheduling priority allocation; Multi-objective optimization solution unit: Supports parallel optimization of multiple objectives such as safety, economy, and environmental protection, and quickly outputs scheduling and trading solutions that take into account multiple dimensions of objectives; Data security encryption transmission unit: Encrypts cross-entity interactive data to ensure data transmission security and prevent information leakage and tampering.

[0021] Preferably, the A2A multi-agent negotiation layer module further includes a dynamic game coordination agent, a health management and maintenance agent, and a data security control agent, and the application service layer module further includes a medium- and long-term scheduling and planning unit, an equipment health and maintenance reporting unit, and a data security audit reporting unit.

[0022] A large-scale multi-agent A2A virtual power grid dispatching and electricity sales method is disclosed. This method, which requires the use of a large-scale multi-agent A2A virtual power grid dispatching and electricity sales process, includes the following steps:

[0023] S1: System Deployment and Agent Registration. Deploy the vertical domain energy big data model module and complete the power corpus fine-tuning to ensure scheduling reasoning and market analysis capabilities. Deploy the MCP tool layer module to complete the encapsulation of tools such as power flow calculation, electricity price forecasting, load forecasting, energy storage SOC estimation, market conditions, meteorological interface, and constraint verification. Deploy the DER agent, scheduling negotiation agent, electricity sales strategy agent, risk assessment agent, review and optimization agent, and optional user preference agent in sequence. Each agent completes registration through the A2A protocol communication unit, publishes the AgentCard, establishes a standardized communication link, and completes system initialization.

[0024] S2: Resource Status Acquisition and Adjustable Capacity Collection Scheduling Negotiation Agent Unit sends the next day's load forecast data, electricity price forecast data, and adjustable capacity collection instructions to each DER agent unit through the A2A protocol communication unit. Each DER agent unit calls the operation monitoring, meteorological interface, and constraint verification unit of the MCP tool layer module to collect the operation parameters of resources such as photovoltaic, wind power, energy storage, gas turbine, flexible load, and charging pile in real time, accurately assess the adjustable power range, duration, and cost function, generate a standardized adjustable capacity report, and send it back to the scheduling negotiation agent unit for aggregation through the A2A protocol.

[0025] S3: Electricity Sales Strategy Formulation and Constraint Synchronization. The electricity sales strategy agent unit calls the market information API unit and electricity price prediction model unit of the MCP tool layer module to obtain and analyze the day-ahead, intraday, and spot market price trends, supply and demand situation, and policy announcements. Combining resource generation costs, energy storage charging and discharging costs, regulation costs, user demand, and competitive situation, it generates differentiated electricity sales pricing strategies, clarifies the pricing range, trading volume, trading period, available electricity volume, and revenue constraints. Through the A2A protocol communication unit, it submits the pricing strategy, revenue constraints, and available electricity volume data to the dispatch negotiation agent unit to complete the alignment of dispatch and electricity sales information.

[0026] S4: Initial Scheme Generation and Distribution, Solicitation of Opinions, and Scheduling Negotiation Agent Unit: The unit summarizes the adjustable capabilities reported by each DER agent and the quotation constraints submitted by the electricity sales strategy agent. Combining grid security constraints and reserve capacity requirements, the unit uses the semantic reasoning unit and decision generation unit of the vertical domain energy big model module to integrate global data to generate an initial scheduling scheme and electricity sales strategy draft that takes into account both grid security and market benefits. The initial scheme is then distributed to DER agents, electricity sales strategy agents, risk assessment agents, and user preference agents through the A2A protocol communication unit to solicit feedback from each agent.

[0027] S5: Multi-round A2A semantic negotiation and conflict resolution. After each agent receives the initial plan, it combines its own responsibilities with feedback and schedules the negotiation agent unit to organize multi-round A2A natural language semantic negotiation, conveying plan details, providing feedback on corrections, and communicating conflict issues. The risk assessment agent unit calls the safety margin monitoring, power limit detection, and power flow constraint verification units in real time to verify the plan's safety and compliance throughout the process, identify risks such as default, price fluctuations, and plan conflicts, and output rejection or correction suggestions. The A2A protocol communication unit automatically identifies negotiation conflicts, assists in resolving contradictions, and gradually converges the plan.

[0028] S6: Consensus Reached and Day-ahead Dispatch Electricity Sales Scheme Output After multiple rounds of negotiation, risk verification and conflict resolution, the dispatch negotiation agent unit integrates all feedback opinions and risk assessment results, and converges the optimal solution through the vertical domain energy big model module to reach a multi-agent consensus. It outputs the final 96-point day-ahead dispatch plan, electricity sales quotation scheme, transaction electricity allocation plan and risk control measures, and distributes them to each agent through the A2A protocol communication unit to notify them to enter the execution phase.

[0029] S7: During the execution of the real-time rolling scheduling and abnormal dynamic adjustment scheme, the DER agent unit monitors the operating status of resource output, energy storage SOC, equipment failure, etc. in real time, and the scheduling negotiation agent unit continuously monitors dynamic information such as market price, load fluctuation, and weather changes. When abnormal scenarios such as sudden changes in new energy output, sudden changes in load, equipment failure, and large price fluctuations occur, A2A emergency negotiation is immediately triggered. The risk assessment agent unit quickly verifies the constraints, and the scheduling negotiation agent unit generates and issues adjustment instructions within minutes to ensure the real-time safety of the power grid and the stability of electricity sales revenue.

[0030] S8: After the execution effect review and strategy closed-loop optimization cycle, the review and optimization agent unit obtains comprehensive data such as scheduling execution data, electricity sales transaction data, actual resource output, market settlement, and deviation assessment through the data acquisition interface in the MCP tool layer module. It evaluates the execution effect from dimensions such as grid security, resource utilization rate, new energy absorption rate, regulation cost, electricity sales revenue, and deviation loss, accurately attributes the causes of deviations, and generates optimization suggestions in combination with the vertical domain energy big model module. These suggestions are fed back to each agent through the A2A protocol communication unit to update the inference strategy and parameter configuration, forming a continuous learning closed loop of execution, evaluation, optimization, and update, and iteratively improving the system's decision-making capabilities.

[0031] Compared with related technologies, the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system and method provided by this invention have the following beneficial effects:

[0032] This invention provides a large-scale multi-agent A2A virtual power grid dispatching and sales system and method. This system introduces standardized A2A protocols into the field of virtual power grid dispatching and sales, achieving standardized interoperability between heterogeneous agents. It supports plug-and-play functionality for agents from different suppliers and with different functions, significantly improving system openness and scalability, reducing system expansion and maintenance costs, filling a technological gap in the field, and realizing integrated coordination of dispatching and sales. It breaks the traditional fragmentation, data incompatibility, and decision-making disconnect between dispatching and sales systems. Through A2A semantic negotiation, it integrates power grid physical security constraints and market revenue objectives, improving the scientific nature and accuracy of decision-making. The large-scale model deeply empowers core decision-making, replacing traditional centralized mathematical optimization models, and possesses policy and weather... The system possesses the ability to understand unstructured information such as market sentiment, enabling it to adapt to complex and dynamic market scenarios, enhance decision-making flexibility, optimize virtual grid resource allocation, significantly improve the renewable energy absorption rate and electricity sales revenue, reduce system operating costs and transaction deviation assessment costs, and effectively enhance overall economic benefits. The intelligent agents interact using natural language, and the entire decision-making process is traceable and explainable, completely solving the black-box decision-making problem of traditional models, improving system operational security and credibility, and constructing a continuous learning closed loop of execution, evaluation, optimization, and updating. Through review and optimization of agents, it iterative optimization of scheduling and electricity sales strategies strengthens the system's adaptive and autonomous evolution capabilities, ensuring long-term stable and efficient operation, and providing a new technical path for the intelligent operation of virtual grids. Attached Figure Description

[0033] Figure 1 A schematic diagram of the first embodiment of the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system provided by the present invention;

[0034] Figure 2 A schematic diagram of a vertical domain energy large model module provided for this invention;

[0035] Figure 3 A schematic diagram of the MCP tool layer module is provided for this invention;

[0036] Figure 4 A schematic diagram of the A2A multi-agent negotiation layer module is provided for this invention;

[0037] Figure 5 A schematic diagram of the application service layer module provided for this invention;

[0038] Figure 6 A schematic diagram of the scheduling negotiation agent unit provided for this invention;

[0039] Figure 7 A schematic diagram of an A2A protocol communication unit is provided for this invention;

[0040] Figure 8This is a schematic diagram of a second embodiment of the large-scale multi-agent A2A virtual power grid dispatching and sales system provided by the present invention;

[0041] Figure 9 A schematic diagram of the dynamic parameter self-calibration unit provided for this invention;

[0042] Figure 10 A schematic diagram of a dynamic game coordination agent is provided for this invention;

[0043] Figure 11 This is a schematic diagram of a medium-to-long-term scheduling planning unit for the present invention. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0045] First Embodiment

[0046] Please refer to the following: Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 and Figure 7 ,in, Figure 1 A schematic diagram of the first embodiment of the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system provided by the present invention; Figure 2 A schematic diagram of a vertical domain energy large model module provided for this invention; Figure 3 A schematic diagram of the MCP tool layer module is provided for this invention; Figure 4 A schematic diagram of the A2A multi-agent negotiation layer module is provided for this invention; Figure 5 A schematic diagram of the application service layer module provided for this invention; Figure 6 A schematic diagram of the scheduling negotiation agent unit provided for this invention; Figure 7 This invention provides a schematic diagram of an A2A protocol communication unit. The large-scale multi-agent A2A virtual power grid dispatching and sales system and method includes: a vertical domain energy large-scale model module, an MCP tool layer module, an A2A multi-agent negotiation layer module, and an application service layer module;

[0047] The vertical domain energy big model module, as the core reasoning engine of the system, relies on the power domain corpus for fine-tuning and has scheduling reasoning and market analysis capabilities. It provides semantic understanding, logical decision-making and unstructured information processing capabilities for each intelligent agent, replacing the traditional mathematical model and supporting adaptive and interpretable scheduling and endogenous power sales decisions.

[0048] The MCP tool layer module, as a standardized tool support layer, uses the MCP protocol to uniformly encapsulate tools such as power flow calculation, load price prediction, and energy storage estimation, and provides a unified calling interface. It does not require customized adaptation code and enables the tools to be used plug and play, providing accurate data and computational support for intelligent agent decision-making.

[0049] The A2A multi-agent negotiation layer module serves as the core of system decision-making. It deploys multiple types of agents and uses the A2A protocol to achieve registration discovery, task delegation, and multi-round semantic negotiation. It coordinates resource adjustability, electricity sales constraints, and security risks to reach consensus on scheduling and electricity sales, supporting heterogeneous resource collaboration and integrated scheduling and trading.

[0050] The application service layer module serves as the external service output layer, integrating the core decision-making results of the system and outputting standardized services such as day-ahead scheduling, real-time adjustment, electricity sales pricing, and post-analysis. It supports visual display and connection with external systems, enabling the implementation of scheduling transaction results and operational decision support.

[0051] In the vertical domain energy big data model module, the corpus in the power field is designed to cover multiple dimensions such as dispatch, market, safety, and users. During the fine-tuning phase, different voltage levels, regional characteristics, and market rules are differentiated. The model inference interface is standardized and encapsulated, and decoupled from the calling logic of each agent. In the MCP tool layer module, the input and output data format specifications of each tool are clearly defined, parameter definitions and error code systems are unified, and new tools complete registration verification in accordance with the MCP protocol. Access permissions and call frequency limits are set for the interfaces. In the A2A multi-agent negotiation layer module, a unified AgentCard format and negotiation process specification are formulated, message priority, timeout mechanism and conflict resolution rules are clarified, and Agent identity authentication and communication encryption mechanisms are established. In the application service layer module, the interface specifications and data interaction protocols for external services are defined, data anonymization and permission isolation strategies are set, and service output meets industry standard format requirements to ensure interoperability with external systems.

[0052] Please refer to Figure 2 The vertical domain energy big model module includes a base big model unit, an electric domain corpus unit, a continuous pre-training unit, a supervised fine-tuning unit, a semantic reasoning unit, and a decision generation unit.

[0053] The base model unit provides fundamental capabilities for general semantic understanding, logical reasoning, and text generation, offering underlying model support for knowledge adaptation in the power sector. This ensures the agent possesses basic natural language interaction and complex task decomposition capabilities. The power sector corpus unit integrates power-specific corpora such as scheduling procedures, market rules, historical transaction data, and equipment operation data, providing high-quality data for domain-specific model training and ensuring the model masters professional knowledge of power scheduling and market transactions. The continuous pre-training unit performs incremental pre-training on the base model based on the power sector corpus, injecting basic power sector knowledge and professional logic to improve the model's understanding of power sector terminology and business processes. The model's understanding depth adapts to the semantic features of power scenarios. The supervised fine-tuning unit performs targeted fine-tuning for core tasks such as scheduling decisions and transaction analysis, enhancing the model's power scheduling knowledge reasoning, market trend analysis, and decision generation capabilities, achieving precise adaptation to both scheduling and transaction scenarios. The semantic reasoning unit possesses the ability to understand unstructured information, breaking through the limitations of traditional models that can only process structured data, supporting adaptive reasoning in complex dynamic scenarios. The decision generation unit directly acts as the intelligent agent's reasoning brain, replacing traditional mathematical optimization models to generate scheduling and electricity sales decisions, eliminating the need for manually constructing a global objective function, and achieving endogenous and intelligent decision output.

[0054] Please refer to Figure 3 The MCP tool layer module includes an MCP protocol encapsulation unit, a power flow calculation solver unit, an electricity price prediction model unit, a load prediction model unit, an energy storage SOC estimator unit, a market information API unit, a meteorological interface unit, a constraint verification unit, and a unified call interface unit.

[0055] MCP Protocol Encapsulation Unit: Based on ModelContextProtocol, this unit builds a standardized encapsulation framework, unifying the calling interfaces and data formats of various tools, eliminating compatibility differences between tools, and enabling plug-and-play functionality. Power Flow Calculation Solver Unit: This unit quickly completes power flow distribution and power balance calculations, accurately verifies the grid security constraints of scheduling schemes, provides quantitative data on grid operation status for intelligent agents, and ensures the physical feasibility of scheduling schemes. Electricity Price Prediction Model Unit: This unit accurately predicts day-ahead, intraday, and spot market electricity price trends, captures price fluctuation patterns, provides price data support for electricity sales strategy formulation and scheduling revenue optimization, and improves the accuracy of trading decisions. Load Prediction Model Unit: This unit accurately predicts the temporal changes in regional electricity load, identifies peak and off-peak periods, provides load benchmark data for resource adjustability planning and scheduling power allocation, and ensures supply and demand balance. Energy Storage SOC Estimator Unit: This unit accurately estimates the remaining capacity of energy storage batteries in real time and dynamically evaluates the charging and discharging capabilities of energy storage. The system features a comprehensive set of functions to optimize energy storage capacity, prevent overcharging and over-discharging, extend energy storage lifespan, and ensure efficient participation of energy storage resources in dispatching. The market data API unit provides real-time dynamic data on electricity market transactions, supply and demand dynamics, and policy announcements, offering the latest market information to the electricity sales strategy agent and supporting dynamic pricing and revenue optimization decisions. The meteorological interface unit acquires real-time meteorological data such as sunlight, wind speed, and temperature, accurately predicting fluctuations in photovoltaic and wind power output, providing meteorological data support for DER adjustable capacity assessment and dispatching scheme adjustments, and reducing the impact of uncertainties in new energy output. The constraint verification unit uniformly verifies multi-dimensional constraints such as power limit violations, power flow violations, equipment operating limits, and market transaction rules, quickly identifying scheme violations and providing constraint basis for intelligent agent negotiation, ensuring scheme compliance. The unified calling interface unit provides a standardized and unique tool calling entry point for upper-layer intelligent agents, simplifying the intelligent agent tool calling logic, eliminating the need to write custom adaptation code for different tools, and reducing system development and maintenance costs.

[0056] Please refer to Figure 4 The A2A multi-agent negotiation layer module includes a DER agent unit, a scheduling negotiation agent unit, a power sales strategy agent unit, a risk assessment agent unit, a user preference agent unit, a review and optimization agent unit, and an A2A protocol communication unit.

[0057] DER Agent Unit: As an intelligent agent for distributed energy resources, it is responsible for collecting operational data from photovoltaic, wind power, and energy storage equipment, assessing adjustable output and response costs, and reporting resource adjustability to the scheduling negotiation agent via the A2A protocol, providing accurate and reliable data support for optimized scheduling. Scheduling Negotiation Agent Unit: As the core hub for multi-agent negotiation, it coordinates resource data, market information, and constraints submitted by various agents, integrates large-scale model reasoning to generate an initial scheduling plan, organizes multiple rounds of semantic negotiation and resolves conflicts, achieving global optimized allocation of virtual grid resources. Electricity Sales Strategy Agent Unit: Focusing on electricity market transactions, it analyzes market conditions, supply and demand dynamics, and competitive landscape, combining resource costs and user needs to formulate differentiated electricity sales pricing strategies and trading plans, coordinating with the scheduling plan for optimization, balancing electricity sales revenue and trading risks. Risk Assessment Agent Unit: Performs full-dimensional security verification of the scheduling plan and trading strategy. Through power flow calculation, constraint verification, and risk modeling, the system identifies potential risks such as power exceeding limits, price fluctuations, and equipment failures, outputting risk levels and correction suggestions to reduce operational and transaction losses. The User Preference Agent unit, acting as an intelligent representative of user needs, collects user preferences such as electricity usage habits, electricity price sensitivity, and reliability requirements, transforming user-side needs into quantifiable constraints. It participates in negotiation to promote solutions that consider user experience and improve user satisfaction. The Post-Analysis and Optimization Agent unit is responsible for the post-event evaluation of the entire scheduling and transaction process, collecting execution data, deviations, and revenue results. It analyzes the effectiveness of the solution execution and the causes of deviations, generates optimization suggestions, and feeds them back to each Agent, driving continuous iterative optimization of the strategy. The A2A Protocol Communication unit provides standardized communication support for multi-agent interaction, enabling Agent registration and discovery, message passing, and task delegation, ensuring the standardization and efficiency of data interaction, supporting plug-and-play expansion, and improving system openness and scalability.

[0058] Please refer to Figure 5 The application service layer module includes a day-ahead scheduling planning unit, a real-time scheduling adjustment unit, an electricity sales price decision unit, a transaction review and analysis unit, a user energy consumption report unit, a visualization display unit, and a standardized interface unit.

[0059] The day-ahead dispatching planning unit outputs a 96-point day-ahead dispatching plan, clearly defining resource output, energy storage charging and discharging, and flexible load response instructions for each time period. This provides a standardized execution basis for the day-ahead operation of the virtual power plant, ensuring the orderly conduct of day-ahead dispatching. The real-time dispatching adjustment unit generates minute-level dispatching adjustment instructions in real time to address abnormal scenarios such as fluctuations in renewable energy output, sudden load changes, equipment failures, and sudden market price changes, ensuring real-time grid security and stable revenue. The electricity sales pricing decision-making unit outputs day-ahead, intraday, and spot market electricity sales pricing schemes, transaction volume allocation plans, and risk control measures, providing standardized decision-making support for electricity sales transactions and improving transaction efficiency and revenue. The transaction review and analysis unit generates a full-link review report for dispatching and transactions. Includes execution results, revenue analysis, deviation attribution, and optimization suggestions, providing data support for operational decisions and facilitating continuous improvement. The user energy consumption report unit generates reports on user energy consumption analysis, cost accounting, green electricity ratio, and optimization suggestions, tailored to user needs, enhancing user service experience and increasing user stickiness. The visualization unit provides a visual display interface for system architecture, resource status, scheduling schemes, transaction data, negotiation processes, risk information, and review results, intuitively presenting the system's operational status and facilitating operation and maintenance and decision-making. The standardized interface unit provides standardized external interfaces, supporting seamless integration with power grid dispatching systems, power market platforms, user-side management systems, and operation and maintenance platforms, improving system compatibility and integrability.

[0060] Please refer to Figure 6 and Figure 7 The scheduling negotiation Agent unit includes a scheduling request receiving unit, a market signal receiving unit, an adjustable capability collection unit, an initial scheme generation unit, a multi-round semantic negotiation unit, a consensus decision-making unit, an A2A message routing unit, and a risk interaction unit. The A2A protocol communication unit also includes an Agent registration unit, an AgentCard management unit, a capability publishing unit, a standardization discovery unit, a task delegation unit, a semantic negotiation unit, a multi-round conversation unit, a conflict resolution unit, a message encoding / decoding unit, and a result feedback unit.

[0061] The dispatch demand receiving unit receives instructions from the power grid dispatch center, including safety constraints, power regulation requirements, and reserve capacity requirements. It clarifies dispatch task objectives and boundary conditions, providing input for dispatch negotiation. The market signal receiving unit receives market signals such as electricity market prices, trading rules, and settlement mechanisms. It integrates dispatch demands with market revenue objectives, supporting integrated dispatch-trading decision-making. The adjustable capacity collection unit broadcasts adjustable capacity collection requests to all DER agents via the A2A protocol, uniformly summarizing adjustable capacity data for each resource and constructing a global resource adjustable capacity view, providing a data foundation for scheme generation. The initial scheme generation unit, based on large-model semantic reasoning, integrates global adjustable capacity, electricity sales constraints, and safety constraints to generate a scheme that takes into account… The initial scheduling scheme and electricity sales strategy draft for power grid security and market revenue are developed as follows: Multi-round semantic negotiation unit: Organizes various agents to conduct A2A multi-round natural language negotiation, conveying scheme details, providing feedback on corrections, and communicating conflict issues, achieving semantic-level information interaction between agents, and gradually converging the scheme; Consensus decision-making unit: Integrates multi-round negotiation feedback and risk assessment results, converges the optimal scheme through large-scale model reasoning, achieves multi-agent consensus, and outputs the final scheduling plan and electricity sales strategy, ensuring the global optimality of the scheme; A2A message routing unit: Coordinates the A2A protocol communication link to achieve accurate distribution, routing, and timing control of negotiation messages, ensuring the orderly progress of the multi-agent negotiation process and avoiding message confusion and loss; Risk interaction unit: Real-time connection with risk assessment A... The agent synchronizes solution information, receives risk verification results, and responds to correction suggestions, controlling solution security risks throughout the process and ensuring compliance and controllability of the negotiation process. The agent registration unit provides a standardized registration entry point for all agents, recording agent identity information, capability information, and communication addresses, establishing a global agent registration ledger to ensure agents can be discovered and identified. The agent card management unit manages the agent cards of each agent, including information such as name, task type, capability description, available tools, and communication endpoints, standardizing agent capability representations and supporting inter-agent capability recognition. The capability publishing unit supports agents publishing their core capabilities, manageable task types, tool access permissions, and other information, achieving global sharing of capability information. The system features several key functionalities: a standardized discovery unit (allowing agents to quickly match collaborating partners), a task delegation unit (supporting standardized task request sending, task allocation, and task authorization between agents, clearly defining task objectives, requirements, time limits, and parameters to ensure clear task delegation and responsibilities), a semantic negotiation unit (supporting the transmission of negotiation content, solution details, and modification opinions in natural language between agents, achieving semantic-level information interaction, overcoming the limitations of traditional numerical communication, and improving negotiation accuracy), and a multi-turn conversation unit (supporting multi-turn negotiation conversations between agents, recording the conversation sequence, content, and status).To ensure the traceability and backtrackability of the negotiation process and support the gradual convergence of complex solutions, the conflict resolution unit automatically identifies negotiation conflicts between agents, provides a conflict coordination mechanism based on A2A protocol semantic rules, assists the risk assessment agent in resolving conflicts, and ensures smooth negotiation progress. The message encoding and decoding unit implements standardized encoding and decoding of A2A protocol messages, unifying message formats, data types, and semantic specifications to ensure unambiguous message transmission and error-free parsing between different agents. The result feedback unit supports the standardized transmission of execution results, negotiation results, and feedback opinions between agents, clearly specifying result status, data details, and supplementary explanations, ensuring a closed-loop information exchange and verifiable results.

[0062] A large-scale multi-agent A2A virtual power grid dispatching and electricity sales method is disclosed. This method, which requires the use of a large-scale multi-agent A2A virtual power grid dispatching and electricity sales process, includes the following steps:

[0063] S1: System Deployment and Agent Registration. Deploy the vertical domain energy big data model module and complete the power corpus fine-tuning to ensure scheduling reasoning and market analysis capabilities. Deploy the MCP tool layer module to complete the encapsulation of tools such as power flow calculation, electricity price forecasting, load forecasting, energy storage SOC estimation, market conditions, meteorological interface, and constraint verification. Deploy the DER agent, scheduling negotiation agent, electricity sales strategy agent, risk assessment agent, review and optimization agent, and optional user preference agent in sequence. Each agent completes registration through the A2A protocol communication unit, publishes the AgentCard, establishes a standardized communication link, and completes system initialization.

[0064] S2: Resource Status Acquisition and Adjustable Capacity Collection Scheduling Negotiation Agent Unit sends the next day's load forecast data, electricity price forecast data, and adjustable capacity collection instructions to each DER agent unit through the A2A protocol communication unit. Each DER agent unit calls the operation monitoring, meteorological interface, and constraint verification unit of the MCP tool layer module to collect the operation parameters of resources such as photovoltaic, wind power, energy storage, gas turbine, flexible load, and charging pile in real time, accurately assess the adjustable power range, duration, and cost function, generate a standardized adjustable capacity report, and send it back to the scheduling negotiation agent unit for aggregation through the A2A protocol.

[0065] S3: Electricity Sales Strategy Formulation and Constraint Synchronization. The electricity sales strategy agent unit calls the market information API unit and electricity price prediction model unit of the MCP tool layer module to obtain and analyze the day-ahead, intraday, and spot market price trends, supply and demand situation, and policy announcements. Combining resource generation costs, energy storage charging and discharging costs, regulation costs, user demand, and competitive situation, it generates differentiated electricity sales pricing strategies, clarifies the pricing range, trading volume, trading period, available electricity volume, and revenue constraints. Through the A2A protocol communication unit, it submits the pricing strategy, revenue constraints, and available electricity volume data to the dispatch negotiation agent unit to complete the alignment of dispatch and electricity sales information.

[0066] S4: Initial Scheme Generation and Distribution, Solicitation of Opinions, and Scheduling Negotiation Agent Unit: The unit summarizes the adjustable capabilities reported by each DER agent and the quotation constraints submitted by the electricity sales strategy agent. Combining grid security constraints and reserve capacity requirements, the unit uses the semantic reasoning unit and decision generation unit of the vertical domain energy big model module to integrate global data to generate an initial scheduling scheme and electricity sales strategy draft that takes into account both grid security and market benefits. The initial scheme is then distributed to DER agents, electricity sales strategy agents, risk assessment agents, and user preference agents through the A2A protocol communication unit to solicit feedback from each agent.

[0067] S5: Multi-round A2A semantic negotiation and conflict resolution. After each agent receives the initial plan, it combines its own responsibilities with feedback and schedules the negotiation agent unit to organize multi-round A2A natural language semantic negotiation, conveying plan details, providing feedback on corrections, and communicating conflict issues. The risk assessment agent unit calls the safety margin monitoring, power limit detection, and power flow constraint verification units in real time to verify the plan's safety and compliance throughout the process, identify risks such as default, price fluctuations, and plan conflicts, and output rejection or correction suggestions. The A2A protocol communication unit automatically identifies negotiation conflicts, assists in resolving contradictions, and gradually converges the plan.

[0068] S6: Consensus Reached and Day-ahead Dispatch Electricity Sales Scheme Output After multiple rounds of negotiation, risk verification and conflict resolution, the dispatch negotiation agent unit integrates all feedback opinions and risk assessment results, and converges the optimal solution through the vertical domain energy big model module to reach a multi-agent consensus. It outputs the final 96-point day-ahead dispatch plan, electricity sales quotation scheme, transaction electricity allocation plan and risk control measures, and distributes them to each agent through the A2A protocol communication unit to notify them to enter the execution phase.

[0069] S7: During the execution of the real-time rolling scheduling and abnormal dynamic adjustment scheme, the DER agent unit monitors the operating status of resource output, energy storage SOC, equipment failure, etc. in real time, and the scheduling negotiation agent unit continuously monitors dynamic information such as market price, load fluctuation, and weather changes. When abnormal scenarios such as sudden changes in new energy output, sudden changes in load, equipment failure, and large price fluctuations occur, A2A emergency negotiation is immediately triggered. The risk assessment agent unit quickly verifies the constraints, and the scheduling negotiation agent unit generates and issues adjustment instructions within minutes to ensure the real-time safety of the power grid and the stability of electricity sales revenue.

[0070] S8: After the execution effect review and strategy closed-loop optimization cycle is completed, the review and optimization agent unit obtains data through the corresponding data acquisition interface in the MCP tool layer module. It comprehensively collects data such as scheduling execution data, electricity sales transaction data, actual resource output, market settlement, and deviation assessment. It evaluates the execution effect from dimensions such as grid security, resource utilization rate, new energy absorption rate, regulation cost, electricity sales revenue, and deviation loss, accurately attributes the cause of deviation, and generates optimization suggestions in combination with the vertical domain energy big model module. These suggestions are fed back to each agent through the A2A protocol communication unit to update the inference strategy and parameter configuration, forming a continuous learning closed loop of execution, evaluation, optimization, and update, and iteratively improving the system's decision-making capabilities.

[0071] The working principle of the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system and method provided by this invention is as follows:

[0072] This system is built on a four-layer architecture based on a large-scale model multi-agent A2A protocol. From bottom to top, it consists of a physical resource layer, an MCP tool layer, an A2A multi-agent negotiation layer, and an application service layer. It comprehensively realizes the distributed resource optimization scheduling of the virtual power grid and the integrated collaborative operation of electricity market sales. The vertical domain energy large-scale model module relies on a general-purpose base model and has been continuously trained and supervised fine-tuned using professional corpora such as power dispatching procedures, market rules, historical transactions, and equipment operation data. It possesses dual capabilities of power dispatching knowledge reasoning and market trend analysis, providing semantic understanding, logical reasoning, and endogenous decision support for each agent within the system, becoming the core reasoning brain for each agent. The MCP tool layer adopts the ModelContextProtocol standard protocol, uniformly encapsulating functional tools such as power flow calculation solvers, electricity price prediction models, load prediction models, energy storage SOC estimators, market information APIs, meteorological interfaces, and constraint checkers. It provides standardized calling interfaces, eliminating the need for customized adaptation code and enabling plug-and-play functionality. It provides accurate and unified data and computational support for agent decision-making. The A2A multi-agent negotiation layer provides the system with... The system centralizes decision-making and deploys DER agents, scheduling negotiation agents, electricity sales strategy agents, risk assessment agents, post-mortem optimization agents, and optional user preference agents. Each agent completes registration and discovery, AgentCard capability release, task delegation and allocation, and multi-round natural language semantic negotiation through the A2A standardized protocol. During system operation, the scheduling negotiation agent collects adjustable capabilities reported by each DER agent through the A2A protocol, while the electricity sales strategy agent simultaneously formulates electricity sales prices and revenue constraints. It integrates large-scale model semantic reasoning to generate initial scheduling and electricity sales plans. After multiple rounds of A2A semantic negotiation, full-process security verification by the risk assessment agent, and conflict resolution, a consensus is reached, and the final scheduling plan and electricity sales strategy are output. During plan execution, resource status, market prices, and other dynamics are monitored in real time and adjusted rapidly. Afterward, the post-mortem optimization agent evaluates the execution effect, attribution deviations, and provides feedback on improvement suggestions, forming a continuous learning closed loop. Finally, the application service layer outputs standardized services such as scheduling, electricity sales, and post-mortem analysis, completing the full-link collaboration of scheduling and electricity sales.

[0073] Compared with related technologies, the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system and method provided by this invention have the following beneficial effects:

[0074] This system introduces the A2A standardized protocol into the field of virtual power grid dispatching and electricity sales, enabling standardized interoperability between heterogeneous intelligent agents. It supports plug-and-play functionality for intelligent agents from different suppliers and with different functions, significantly improving system openness and scalability, reducing system expansion and maintenance costs, filling technological gaps in the field, and achieving integrated coordination of dispatching and electricity sales. This breaks down the traditional fragmentation, data barriers, and disconnected decision-making between dispatching and electricity sales systems. Through A2A semantic negotiation, it integrates power grid physical security constraints and market revenue objectives, improving the scientific rigor and accuracy of decision-making. Large-scale models deeply empower core decision-making, replacing traditional centralized mathematical optimization models, and possessing the ability to understand unstructured information such as policies, weather, and market sentiment. This system possesses the capability to adapt to complex and dynamic market scenarios, enhancing decision-making flexibility, optimizing virtual grid resource allocation, significantly improving renewable energy absorption and electricity sales revenue, reducing system operating costs and transaction deviation assessment costs, and effectively enhancing overall economic benefits. Intelligent agents interact using natural language, and the entire decision-making process is traceable and explainable, completely solving the black-box decision-making problem of traditional models, improving system operational security and credibility, and constructing a continuous learning closed loop of execution, evaluation, optimization, and updating. Through review and optimization of agents, it iterative optimization of scheduling and electricity sales strategies strengthens the system's adaptive and autonomous evolution capabilities, ensuring long-term stable and efficient operation, and providing a new technical path for the intelligent operation of virtual grids.

[0075] Second Embodiment

[0076] Please refer to the following: Figures 8-9 - Figures 10-11 , Figure 8 This is a schematic diagram of a second embodiment of the large-scale multi-agent A2A virtual power grid dispatching and sales system provided by the present invention; Figure 9 A schematic diagram of the dynamic parameter self-calibration unit provided for this invention; Figure 10 A schematic diagram of a dynamic game coordination agent is provided for this invention; Figure 11 This invention provides a schematic diagram of a medium-to-long-term scheduling planning unit. Based on the large-scale multi-agent A2A virtual power grid scheduling and sales system and method provided in the first embodiment of this application, the second embodiment of this application proposes another large-scale multi-agent A2A virtual power grid scheduling and sales system and method. The second embodiment is merely a preferred embodiment of the first embodiment, and the implementation of the second embodiment will not affect the separate implementation of the first embodiment.

[0077] Specifically, the difference between the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system and method provided in the second embodiment of this application is as follows: Please refer to... Figure 8 The vertical energy big model module also includes a dynamic scenario adaptive reasoning unit, a multi-agent game equilibrium reasoning unit, and a long-cycle trend inference unit.

[0078] Dynamic Scenario Adaptive Reasoning Unit: For dynamic scenarios such as strong fluctuations in new energy sources, sudden changes in load, and anomalies in grid topology, it adjusts the reasoning logic and decision-making strategies in real time to enhance the decision-making adaptability under complex operating conditions. Multi-Agent Game Equilibrium Reasoning Unit: Simulates the game relationship between multiple virtual power plants, grids, and users, and reasones equilibrium decision schemes to take into account the interests of all parties and ensure the stability of collaborative operation. Long-Term Trend Inference Unit: Based on historical data and industry policies, it infers the medium- and long-term trends of resource output, market prices, and load changes to support medium- and long-term dispatching and trading planning.

[0079] Please refer to Figure 9 The MCP tool layer module also includes a dynamic parameter self-calibration unit, a distributed resource health assessment unit, a multi-objective optimization solution unit, and a data security encrypted transmission unit.

[0080] Dynamic parameter self-calibration unit: Real-time calibration of load, electricity price, and new energy output prediction model parameters, correcting prediction deviations and improving data accuracy in dynamic scenarios; Distributed resource health assessment unit: Real-time monitoring of equipment health status, wear level, and aging trend, quantifying resource availability, and providing a basis for scheduling priority allocation; Multi-objective optimization solution unit: Supports parallel optimization of multiple objectives such as safety, economy, and environmental protection, and quickly outputs scheduling and trading solutions that take into account multiple dimensions of objectives; Data security encryption transmission unit: Encrypts cross-entity interactive data to ensure data transmission security and prevent information leakage and tampering.

[0081] Please refer to Figure 10 and Figure 11 The A2A multi-agent negotiation layer module also includes a dynamic game coordination agent, a health management and maintenance agent, and a data security control agent. The application service layer module also includes a medium- and long-term scheduling planning unit, an equipment health and maintenance reporting unit, and a data security audit reporting unit.

[0082] Dynamic Game Coordination Agent: Coordinates the game relationships among multiple stakeholders, harmonizes their interests, resolves conflicting objectives, and promotes balanced decisions that consider security, profitability, and fairness. Health Management and Maintenance Agent: Responsible for distributed resource health monitoring, fault early warning, and maintenance task scheduling to ensure reliable resource operation and extend equipment lifespan. Data Security Control Agent: Responsible for cross-stakeholder data interaction permission management, encryption verification, and security auditing to ensure system data security and compliance and prevent data risks. New Medium- and Long-Term Scheduling Planning Unit: Outputs weekly / monthly / quarterly medium- and long-term resource allocation, maintenance plans, and transaction plans to support the long-term stable operation of the virtual power grid. Equipment Health and Maintenance Reporting Unit: Generates resource health status, fault early warning, and maintenance suggestion reports to support full lifecycle management of equipment. Data Security Audit Reporting Unit: Outputs data interaction logs, permission usage records, and security risk reports to support data security compliance control.

[0083] Compared with related technologies, the large-scale multi-agent A2A virtual power grid dispatching and electricity sales system and method provided by this invention have the following beneficial effects:

[0084] The existing system is enhanced by adding dynamic adaptation, game equilibrium, health maintenance, and data security-related units and agents, constructing an enhanced system that is dynamically adaptive, multi-stakeholder balanced, health-maintainable, and security-controllable. The vertical domain energy big data model adds adaptive, game equilibrium, and long-cycle extrapolation units to strengthen dynamic scenario adaptation, multi-stakeholder interest balancing, and medium-to-long-term planning capabilities. The MCP tool layer adds parameter calibration, health assessment, multi-objective optimization, and data encryption units to improve data accuracy, resource reliability, multi-objective solution capabilities, and data security. The A2A negotiation layer adds game coordination, health maintenance, and data security agents. By leveraging the A2A protocol to achieve multi-stakeholder game equilibrium, resource health management, and data security control interaction, this implementation coordinates dynamic adaptation, interest balancing, operation and maintenance assurance, and security compliance. The application service layer adds medium- and long-term planning, operation and maintenance reports, and security audit units, outputting long-term operation, equipment management, and security control services. This embodiment enhances the system's dynamic adaptability, making it more robust in complex and fluctuating scenarios, balancing the interests of multiple stakeholders, improving the stability of collaborative operation, strengthening equipment health management, reducing the risk of failure and outages, ensuring data security compliance, and preventing information leakage risks. It provides comprehensive technical support for the long-term, safe, efficient, and collaborative operation of the virtual power grid.

[0085] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A large-scale multi-agent A2A virtual power grid dispatching and electricity sales system, characterized in that, include: Vertical domain energy big model module, MCP tool layer module, A2A multi-agent negotiation layer module and application service layer module; The vertical domain energy big model module, as the core reasoning engine of the system, relies on the power domain corpus for fine-tuning and has scheduling reasoning and market analysis capabilities. It provides semantic understanding, logical decision-making and unstructured information processing capabilities for each intelligent agent, replacing the traditional mathematical model and supporting adaptive and interpretable scheduling and endogenous power sales decisions. The MCP tool layer module, as a standardized tool support layer, uses the MCP protocol to uniformly encapsulate tools such as power flow calculation, load price prediction, and energy storage estimation, and provides a unified calling interface. It does not require customized adaptation code and enables the tools to be used plug and play, providing accurate data and computational support for intelligent agent decision-making. The A2A multi-agent negotiation layer module serves as the core of system decision-making. It deploys multiple types of agents and uses the A2A protocol to achieve registration discovery, task delegation, and multi-round semantic negotiation. It coordinates resource adjustability, electricity sales constraints, and security risks to reach consensus on scheduling and electricity sales, supporting heterogeneous resource collaboration and integrated scheduling and trading. The application service layer module serves as the external service output layer. It integrates the core decision-making results of the system and outputs standardized services such as day-ahead scheduling, real-time adjustment, electricity sales pricing, and retrospective analysis. It supports visual display and interface with external systems, enabling the implementation of scheduling transaction results and operational decision support.

2. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The vertical domain energy big model module includes a base big model unit, an electric domain corpus unit, a continuous pre-training unit, a supervised fine-tuning unit, a semantic reasoning unit, and a decision generation unit.

3. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The MCP tool layer module includes an MCP protocol encapsulation unit, a power flow calculation solver unit, an electricity price prediction model unit, a load prediction model unit, an energy storage SOC estimator unit, a market information API unit, a meteorological interface unit, a constraint verification unit, and a unified call interface unit.

4. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The A2A multi-agent negotiation layer module includes a DER agent unit, a scheduling negotiation agent unit, a power sales strategy agent unit, a risk assessment agent unit, a user preference agent unit, a review and optimization agent unit, and an A2A protocol communication unit.

5. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The application service layer module includes a day-ahead scheduling planning unit, a real-time scheduling adjustment unit, an electricity sales pricing decision unit, a transaction review and analysis unit, a user energy consumption report unit, a visualization display unit, and a standardized interface unit.

6. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The scheduling negotiation agent unit includes a scheduling request receiving unit, a market signal receiving unit, an adjustable capability collection unit, an initial scheme generation unit, a multi-round semantic negotiation unit, a consensus decision-making unit, an A2A message routing unit, and a risk interaction unit. The A2A protocol communication unit also includes an agent registration unit, an agent card management unit, a capability publishing unit, a standardization discovery unit, a task delegation unit, a semantic negotiation unit, a multi-round conversation unit, a conflict resolution unit, a message encoding / decoding unit, and a result feedback unit.

7. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The vertical energy big model module also includes a dynamic scenario adaptive reasoning unit, a multi-agent game equilibrium reasoning unit, and a long-term trend inference unit.

8. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The MCP tool layer module also includes a dynamic parameter self-calibration unit, a distributed resource health assessment unit, a multi-objective optimization solution unit, and a data security encrypted transmission unit.

9. The large-scale multi-agent A2A virtual power grid dispatching and sales system according to claim 1, characterized in that, The A2A multi-agent negotiation layer module also includes a dynamic game coordination agent, a health management and maintenance agent, and a data security control agent. The application service layer module also includes a medium- and long-term scheduling and planning unit, an equipment health and maintenance reporting unit, and a data security audit reporting unit.

10. A large-scale multi-agent A2A virtual power grid dispatching and electricity sales method, characterized in that, The system includes a large-scale multi-agent A2A virtual power grid dispatching and sales system as described in any one of claims 1-9. The large-scale multi-agent A2A virtual power grid dispatching and sales system requires the use of a large-scale multi-agent A2A virtual power grid dispatching and sales method during power sales, comprising the following steps: S1: System Deployment and Agent Registration. Deploy the vertical domain energy big data model module and complete the power corpus fine-tuning to ensure scheduling reasoning and market analysis capabilities. Deploy the MCP tool layer module to complete the encapsulation of tools such as power flow calculation, electricity price forecasting, load forecasting, energy storage SOC estimation, market conditions, meteorological interface, and constraint verification. Deploy the DER agent, scheduling negotiation agent, electricity sales strategy agent, risk assessment agent, review and optimization agent, and optional user preference agent in sequence. Each agent completes registration through the A2A protocol communication unit, publishes the AgentCard, establishes a standardized communication link, and completes system initialization. S2: Resource Status Acquisition and Adjustable Capacity Collection Scheduling Negotiation Agent Unit sends the next day's load forecast data, electricity price forecast data, and adjustable capacity collection instructions to each DER agent unit through the A2A protocol communication unit. Each DER agent unit calls the operation monitoring, meteorological interface, and constraint verification unit of the MCP tool layer module to collect the operation parameters of resources such as photovoltaic, wind power, energy storage, gas turbine, flexible load, and charging pile in real time, accurately assess the adjustable power range, duration, and cost function, generate a standardized adjustable capacity report, and send it back to the scheduling negotiation agent unit for aggregation through the A2A protocol. S3: Electricity Sales Strategy Formulation and Constraint Synchronization. The electricity sales strategy agent unit calls the market information API unit and electricity price prediction model unit of the MCP tool layer module to obtain and analyze the day-ahead, intraday, and spot market price trends, supply and demand situation, and policy announcements. Combining resource generation costs, energy storage charging and discharging costs, regulation costs, user demand, and competitive situation, it generates differentiated electricity sales pricing strategies, clarifies the pricing range, trading volume, trading period, available electricity volume, and revenue constraints. Through the A2A protocol communication unit, it submits the pricing strategy, revenue constraints, and available electricity volume data to the dispatch negotiation agent unit to complete the alignment of dispatch and electricity sales information. S4: Initial Scheme Generation and Distribution, Solicitation of Opinions, and Scheduling Negotiation Agent Unit: The unit summarizes the adjustable capabilities reported by each DER agent and the quotation constraints submitted by the electricity sales strategy agent. Combining grid security constraints and reserve capacity requirements, the unit uses the semantic reasoning unit and decision generation unit of the vertical domain energy big model module to integrate global data to generate an initial scheduling scheme and electricity sales strategy draft that takes into account both grid security and market benefits. The initial scheme is then distributed to DER agents, electricity sales strategy agents, risk assessment agents, and user preference agents through the A2A protocol communication unit to solicit feedback from each agent. S5: Multi-round A2A semantic negotiation and conflict resolution. After each agent receives the initial plan, it combines its own responsibilities with feedback and schedules the negotiation agent unit to organize multi-round A2A natural language semantic negotiation, conveying plan details, providing feedback on corrections, and communicating conflict issues. The risk assessment agent unit calls the safety margin monitoring, power limit detection, and power flow constraint verification units in real time to verify the plan's safety and compliance throughout the process, identify risks such as default, price fluctuations, and plan conflicts, and output rejection or correction suggestions. The A2A protocol communication unit automatically identifies negotiation conflicts, assists in resolving contradictions, and gradually converges the plan. S6: Consensus Reached and Day-ahead Dispatch Electricity Sales Scheme Output After multiple rounds of negotiation, risk verification and conflict resolution, the dispatch negotiation agent unit integrates all feedback opinions and risk assessment results, and converges the optimal solution through the vertical domain energy big model module to reach a multi-agent consensus. It outputs the final 96-point day-ahead dispatch plan, electricity sales quotation scheme, transaction electricity allocation plan and risk control measures, and distributes them to each agent through the A2A protocol communication unit to notify them to enter the execution phase. S7: During the execution of the real-time rolling scheduling and abnormal dynamic adjustment scheme, the DER agent unit monitors the operating status of resource output, energy storage SOC, equipment failure, etc. in real time, and the scheduling negotiation agent unit continuously monitors dynamic information such as market price, load fluctuation, and weather changes. When abnormal scenarios such as sudden changes in new energy output, sudden changes in load, equipment failure, and large price fluctuations occur, A2A emergency negotiation is immediately triggered. The risk assessment agent unit quickly verifies the constraints, and the scheduling negotiation agent unit generates and issues adjustment instructions within minutes to ensure the real-time safety of the power grid and the stability of electricity sales revenue. S8: After the execution effect review and strategy closed-loop optimization cycle, the review and optimization agent unit obtains comprehensive data such as scheduling execution data, electricity sales transaction data, actual resource output, market settlement, and deviation assessment through the data acquisition interface in the MCP tool layer module. It evaluates the execution effect from dimensions such as grid security, resource utilization rate, new energy absorption rate, regulation cost, electricity sales revenue, and deviation loss, accurately attributes the causes of deviations, and generates optimization suggestions in combination with the vertical domain energy big model module. These suggestions are fed back to each agent through the A2A protocol communication unit to update the inference strategy and parameter configuration, forming a continuous learning closed loop of execution, evaluation, optimization, and update, and iteratively improving the system's decision-making capabilities.

Images