A Multi-Agent Cooperative Decision-Making and Task Decomposition Method and System for Power Dispatch

Abstract

This invention discloses a multi-agent collaborative decision-making and task decomposition method and system for power dispatching, belonging to the fields of intelligent power system dispatching and artificial intelligence technology. The method constructs a power system operating state space based on a unified state object model. Through a task parsing module, it identifies dispatching objectives and decomposes dispatching tasks into load forecasting sub-tasks, energy storage optimization sub-tasks, safety verification sub-tasks, carbon optimization sub-tasks, and dispatching generation sub-tasks. These sub-tasks are collaboratively decided by the forecasting agent, energy storage agent, safety agent, carbon optimization agent, and dispatching agent, respectively. A collaborative decision-making engine performs conflict detection, consistency verification, and result fusion on the output results of each agent to generate the final dispatching scheme. The scheme is then output to the dispatching execution system through a standardized service interface. This invention achieves automatic decomposition of complex dispatching tasks, multi-agent collaborative reasoning, and joint decision-making, improving the efficiency and intelligence level of power system dispatching decisions.

Description

Technical Field

[0001] This invention belongs to the field of intelligent dispatching and artificial intelligence technology of power systems, specifically involving a multi-agent collaborative decision-making and task decomposition method and system for power dispatching, applicable to scenarios such as power grid dispatching centers, virtual power plants, zero-carbon parks and integrated energy systems. Background Technology

[0002] With the deepening of the construction of new power systems, the large-scale grid connection of new energy sources, the widespread access of distributed power sources, and the large-scale deployment of energy storage systems, the uncertainty of power system operation has increased significantly, and the complexity and real-time requirements of dispatching decisions have continued to rise. Traditional dispatching methods mainly rely on dispatcher experience and established procedures, which make it difficult to make optimal decisions quickly when faced with complex dispatching scenarios with multiple objectives, multiple constraints, and multiple time scales.

[0003] In recent years, artificial intelligence technology has received widespread attention in the field of power system dispatching. Existing research has applied methods such as deep learning and reinforcement learning to tasks including load forecasting, economic dispatching, and security assessment. However, current technologies still have the following shortcomings: (1) The scheduling task is singular. Existing methods are usually designed for a single scheduling objective (such as optimal economy or priority of safety), and it is difficult to simultaneously take into account the coordinated optimization of multiple objectives such as economy, safety, and low carbon. (2) Lack of coordination mechanism in decision-making process. Different scheduling tasks (such as load forecasting, energy storage optimization, and safety verification) are usually completed by independent modules. There is no effective coordination decision-making mechanism between the modules, making it difficult to form a unified scheduling scheme. (3) Task decomposition relies on human experience. The decomposition and execution of complex scheduling tasks heavily depend on the scheduler's experience, making it difficult to achieve automated and intelligent decomposition and allocation. (4) Multi-agent collaboration has not yet formed an effective framework. Although some existing technologies involve multi-agent methods, there is a lack of systematic solutions for multi-agent collaborative decision-making and task decomposition for power dispatching scenarios.

[0004] In recent years, AI agent technology has made significant breakthroughs in autonomous decision-making and task execution in general fields, providing a new technological path for the intelligent upgrading of the power dispatching field. Therefore, there is an urgent need for a power dispatching method and system that can automatically decompose complex dispatching tasks and enable multiple AI agents to collaboratively complete decision-making. Summary of the Invention

[0005] Purpose of the invention The purpose of this invention is to provide a multi-agent collaborative decision-making and task decomposition method and system for power dispatching, so as to solve the technical problems of single dispatching tasks, lack of collaborative mechanism in decision-making process, and reliance on human experience in task decomposition in the prior art.

[0006] Technical solution To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a multi-agent cooperative decision-making and task decomposition method for power dispatching, comprising the following steps:

[0007] Step S1: Acquire power system operating status data and construct a unified state object model. The unified state object model includes state attributes, topology attributes, time attributes, reliability attributes, data source attributes, and relationship attributes. Specifically, state attributes characterize the power system operating state vector; topology attributes characterize the connection relationships between power grid nodes; time attributes characterize the timestamp and timeliness of the data; reliability attributes characterize the confidence level or source credibility level of the data; data source attributes identify the data source system; and relationship attributes characterize the correlation relationships between data.

[0008] Step S2: Analyze the current scheduling task based on the scheduling objectives. The scheduling objectives include at least one of the following: economic objectives, safety objectives, low-carbon emission objectives, and renewable energy consumption objectives. The scheduling tasks include one or more of the following: economic scheduling tasks, safety-constrained scheduling tasks, energy storage optimization tasks, renewable energy consumption tasks, carbon optimization tasks, and load control tasks.

[0009] Step S3: Decompose the scheduling task into multiple subtasks. The subtasks include load prediction subtasks, energy storage optimization subtasks, safety verification subtasks, carbon optimization subtasks, and scheduling generation subtasks. The task decomposition is based on the scheduling objective and the current system state, and uses a task tree structure to decompose the composite scheduling task layer by layer into atomic-level subtasks; the task tree structure includes parent task nodes, child task nodes, and dependency edges.

[0010] Step S4: Assign each subtask to the corresponding intelligent agent for execution. Based on the subtask type, task dependencies, resource status, and agent capabilities, each subtask is dynamically assigned to the corresponding intelligent agent. The intelligent agents include a prediction agent, an energy storage agent, a safety agent, a carbon optimization agent, and a scheduling agent. The prediction agent is used to predict future load and renewable energy output; the energy storage agent is used to generate energy storage charging and discharging strategies; the safety agent is used to verify grid safety constraints; the carbon optimization agent is used to generate low-carbon operation schemes; and the scheduling agent is used to synthesize the output results of each intelligent agent to generate a scheduling scheme.

[0011] Step S5: The collaborative decision-making engine performs collaborative decision-making and result fusion on the output results of each agent. The collaborative decision-making includes conflict detection, consistency verification, priority coordination, and result fusion. Conflict detection identifies contradictions or exceedances in the output results of each agent; consistency verification verifies whether each output result conforms to the constraints of the unified state object model; priority coordination weights or sorts the outputs of each agent according to the priority of the scheduling target; result fusion uses one or more of the following methods: weighted fusion, rule fusion, game fusion, negotiation fusion, and voting fusion, to generate a fused decision result.

[0012] Step S6: Generate the final scheduling scheme and output it to the scheduling execution system through a standardized service interface. The standardized service interface is compatible with the unified state object model interface specification and can directly interface with the scheduling execution system or the intelligent agent module.

[0013] Furthermore, the method supports at least one of the following scheduling modes: economic scheduling priority mode, safety scheduling priority mode, low-carbon scheduling priority mode, new energy consumption maximization mode, and multi-objective equilibrium mode.

[0014] Secondly, the present invention provides a multi-agent cooperative decision-making system for power dispatching, comprising:

[0015] The system comprises the following modules: a state modeling module for acquiring power system operating state data and constructing a unified state object model; a task parsing module for parsing the current scheduling task based on the scheduling objective; a task decomposition module for decomposing the scheduling task into multiple sub-tasks; a task allocation module for dynamically allocating sub-tasks to corresponding agents based on sub-task type, task dependencies, resource status, and agent capabilities; an agent coordination module, including prediction agents, energy storage agents, security agents, carbon optimization agents, and scheduling agents, for executing their respective sub-tasks and outputting results; a collaborative decision-making module for performing conflict detection, consistency verification, priority coordination, and result fusion on the output results of each agent to generate the final scheduling scheme; and a scheduling output module for outputting the final scheduling scheme to the scheduling execution system through a standardized service interface.

[0016] Beneficial effects Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) Automatic decomposition of complex scheduling tasks. Through task parsing and decomposition mechanism, complex scheduling tasks are automatically decomposed into atomic-level subtasks, reducing reliance on human experience and improving the efficiency and accuracy of scheduling task processing.

[0018] (2) Multi-agent collaborative decision-making. By constructing a collaborative decision-making framework of predictive agent, energy storage agent, safety agent, carbon optimization agent and scheduling agent, distributed intelligent decision-making under multi-objective and multi-constraint scenarios can be realized, effectively improving the global optimality of scheduling schemes.

[0019] (3) Conflict detection and consistency verification. The collaborative decision engine performs conflict detection, consistency verification and priority coordination on the output results of multiple agents to ensure that the scheduling scheme meets the power grid security constraints and operational reliability requirements.

[0020] (4) Multi-mode scheduling support. Supports multiple scheduling modes such as economic scheduling priority, safety scheduling priority, low-carbon scheduling priority, maximizing new energy consumption and multi-objective balance, to adapt to the differentiated needs under different operating scenarios.

[0021] (5) Compatible with relevant patent systems. This invention can work in conjunction with the unified state object model construction module, service interface module, data middle platform module, and capability orchestration module to form a complete closed loop from data access to scheduling decision output. Attached Figure Description

[0023] Figure 1 This is an overall flowchart of the method of the present invention.

[0024] Figure 2 This is a schematic diagram of the multi-agent collaborative decision-making architecture of the present invention.

[0025] Figure 3 This is a schematic diagram of the task decomposition and agent allocation structure of the present invention.

[0026] Figure 4 This is a flowchart of the collaborative decision engine of the present invention. Detailed Implementation

[0027] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0028] Example 1: Multi-agent cooperative decision-making process like Figure 1 As shown in the figure, this embodiment provides a multi-agent collaborative decision-making and task decomposition method for power dispatch, with the goal of maximizing the consumption of new energy sources. The complete process is as follows:

[0029] S1: The state modeling module obtains power grid operation data from the SCADA system and PMU device, and constructs a unified state object model, which includes state attributes such as node voltage, line power flow, generator output, load power, energy storage SOC, and new energy output, as well as attributes such as power grid topology, timestamp, and data reliability.

[0030] S2: The task parsing module receives the scheduling objective of "maximizing the consumption of new energy", parses it into a new energy consumption task, and identifies the types of sub-tasks involved.

[0031] S3: The task decomposition module decomposes the scheduling task into five sub-tasks: load forecasting sub-task (predicting the load curve for the next 24 hours or the load curve for the next 96 15-minute scheduling cycles), energy storage optimization sub-task (formulating energy storage charging and discharging plans), safety verification sub-task (verifying grid safety constraints), carbon optimization sub-task (assessing the impact of carbon emissions), and scheduling generation sub-task (generating the final scheduling plan).

[0032] S4: The task allocation module assigns the load prediction subtask to the prediction agent, the energy storage optimization subtask to the energy storage agent, the safety verification subtask to the safety agent, the carbon optimization subtask to the carbon optimization agent, and the scheduling generation subtask to the scheduling agent based on the type of each subtask, task dependency, resource status, and agent capabilities.

[0033] S5: Each agent executes its corresponding sub-task in parallel. The prediction agent uses one or more of the following: machine learning models, deep learning models, time-series prediction models, or large-scale models, to predict the load curve and renewable energy output curve for the next 24 hours based on historical load data and meteorological data; the energy storage agent generates an energy storage charging and discharging strategy based on the prediction results and electricity price signals; the safety agent calls the power flow calculation program to verify whether the node voltage and line power flow under each candidate scheme exceed the limits; the carbon optimization agent generates a low-carbon operation adjustment strategy based on carbon emission intensity prediction and carbon price; and the scheduling agent receives the output results from each agent and performs a comprehensive evaluation.

[0034] S6: The collaborative decision engine performs conflict detection (such as whether there is a conflict between the energy storage scheme and the safety verification result), consistency verification (whether the constraints of the unified state object model are met), and priority coordination (coordinating according to preset priority rules, such as: safety first, low carbon second, economy third) on the output results of each intelligent agent, and generates the final scheduling scheme by one or more of the following methods: weighted fusion, rule fusion, game fusion, negotiation fusion or voting fusion.

[0035] S7: The scheduling output module outputs the final scheduling scheme to the scheduling execution system through a standardized service interface.

[0036] Example 2: Multi-agent cooperative decision-making architecture

[0037] like Figure 2As shown, the multi-agent collaborative decision-making system architecture provided in this embodiment includes: a state modeling module, a task parsing module, a task decomposition module, a task allocation module, an agent collaboration module (including a predictive agent, an energy storage agent, a safety agent, a carbon optimization agent, and a scheduling agent), a collaborative decision-making module, and a scheduling output module.

[0038] The state modeling module is compatible with the unified state object model building module, and the input / output format of the agent collaboration module is compatible with the data platform and capability orchestration module. Communication between agents uses a standardized message format and supports both synchronous and asynchronous communication modes.

[0039] Example 3: Multi-mode scheduling switching

[0040] This embodiment supports switching between multiple scheduling modes. When the system detects that the power grid is under peak load, it automatically switches to the "safety scheduling priority mode" and increases the output weight of the safety agent to the highest level. When the system is generating new energy sources and the power grid load is low, it automatically switches to the "new energy consumption maximization mode" and increases the output weight of the prediction agent and the energy storage agent. When the system is operating at a high carbon price, it automatically switches to the "low-carbon scheduling priority mode" and increases the output weight of the carbon optimization agent.

[0041] Example 4: Conflict Handling Mechanism of Collaborative Decision Engine In this embodiment, the conflict resolution process of the collaborative decision-making engine for the output results of multiple agents is as follows:

[0042] The first step is conflict detection: comparing the outputs of each agent to determine if there are any conflicts related to objectives, constraints, or resources. If conflicts exist, they are marked as conflict items. Among them, objective conflicts include conflicts between the low-carbon operation plan generated by the carbon optimization agent and the low-cost operation plan generated by the economic scheduling agent; constraint conflicts include situations where the charging and discharging plan output by the energy storage agent leads to the safety agent detecting line overload, voltage exceeding limits, etc.

[0043] The second step is consistency verification: verifying whether the output results of each intelligent agent meet the constraints of the unified state object model, including node voltage constraints, line power flow constraints, energy storage SOC constraints, and equipment operation boundary constraints.

[0044] The third step is priority coordination: the outputs of each agent are sorted and weighted according to the preset priority rules, with safety constraints as hard constraints being given priority to be met, and economic and low-carbon objectives as soft constraints being coordinated and optimized.

[0045] The fourth step is result fusion: using one or more of the following methods, namely negotiation fusion, rule fusion, weighted fusion, or game fusion, the output results of the agent after priority coordination are fused to generate a comprehensive scheduling scheme that meets security constraints and takes into account economy and low carbon emissions.

[0046] Example 5: Synergistic application with relevant patent systems This embodiment demonstrates the collaborative application of the multi-agent collaborative decision-making method with the unified state object model construction module, service interface module, data middleware platform module, and capability orchestration module: The unified state object model construction module provides standardized data input for this invention; the data middle platform module provides data management and service support for this invention; the capability orchestration module provides dynamic invocation and orchestration services for scheduling capabilities for this invention; the multi-agent collaborative decision output results of this invention are returned to the scheduling execution system through a standardized service interface, forming a complete closed loop from data access to scheduling decision output.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-agent collaborative decision-making and task decomposition method for power dispatching, characterized in that, Includes the following steps: S1: Acquire power system operation status data and construct a unified state object model; the unified state object model includes at least state attributes, topology attributes, time attributes, reliability attributes, data source attributes, and relationship attributes; S2: Parse the current scheduled task based on the scheduling target; S3: Decompose the scheduling task into multiple sub-tasks, including load forecasting sub-task, energy storage optimization sub-task, safety verification sub-task, carbon optimization sub-task and scheduling generation sub-task; S4: Assign each subtask to the corresponding intelligent agent for execution, the intelligent agents including prediction intelligent agent, energy storage intelligent agent, safety intelligent agent, carbon optimization intelligent agent and scheduling intelligent agent; S5: The collaborative decision-making engine performs collaborative decision-making and result fusion on the output results of each agent. The collaborative decision-making includes conflict detection, consistency verification, priority coordination and result fusion. S6: Generate the final scheduling scheme and output it to the scheduling execution system through a standardized service interface.

2. The method according to claim 1, characterized in that, The state attributes in the unified state object model are used to characterize the power system operating state vector; the topology attributes are used to characterize the grid node connection relationship; and the time attributes are used to characterize the timestamp and timeliness of the data. The confidence level attribute is used to characterize the confidence level of the data or the confidence level of the source; the data source attribute is used to identify the data source system. Relationship attributes are used to characterize the relationships between data.

3. The method according to claim 1, characterized in that, The scheduling tasks include one or more of the following: economic scheduling tasks, safety-constrained scheduling tasks, energy storage optimization tasks, new energy consumption tasks, carbon optimization tasks, and load control tasks; the task decomposition adopts a task tree structure to decompose the composite scheduling tasks into atomic-level sub-tasks layer by layer, and the task tree structure includes parent task nodes, child task nodes, and dependency edges.

4. The method according to claim 1, characterized in that, The predictive agent employs one or more of the following: machine learning models, deep learning models, time-series prediction models, or large-scale models. Its inputs are historical load data and renewable energy output data, and its outputs are load forecast curves and renewable energy output forecast curves within a future time window. The energy storage agent's inputs are energy storage state of charge, charging and discharging efficiency, and electricity price signals, and its output is an energy storage charging and discharging power plan. The safety agent's inputs are grid topology and power flow data, and its outputs are safety constraint overrun detection results and feasibility assessments. The carbon optimization agent's inputs are carbon emission intensity data, carbon price, and load forecast results, and its output is a low-carbon operation adjustment strategy. The input to the scheduling agent is the output of other agents and the fusion weights of the collaborative decision engine, and the output is a comprehensive scheduling scheme.

5. The method according to claim 1, characterized in that, The conflict detection is used to identify contradictions or exceedances in the output results of each agent; the consistency verification is used to verify whether each output result conforms to the constraints of the unified state object model; the priority coordination is used to weight or sort the outputs of each agent according to the priority of the scheduling target.

6. The method according to claim 1, characterized in that, The results fusion adopts one or more of the following methods: weighted fusion, rule fusion, game fusion, negotiation fusion, or voting fusion.

7. The method according to claim 1, characterized in that, The method supports at least one of the following scheduling modes: economic scheduling priority mode, safety scheduling priority mode, low-carbon scheduling priority mode, new energy consumption maximization mode, and multi-objective equilibrium mode.

8. The method according to claim 1, characterized in that, The method further includes a task allocation step: after task decomposition and before agent execution, each subtask is dynamically allocated to the corresponding agent based on the subtask type, task dependency, resource status, and agent capabilities.

9. A multi-agent cooperative decision-making system for power dispatching, used to implement the method described in any one of claims 1 to 8, characterized in that, include: The state modeling module is used to build a unified state object model; The task parsing module is used to parse the currently scheduled task based on the scheduling target; The task decomposition module is used to decompose a scheduled task into multiple subtasks; The task allocation module is used to dynamically allocate subtasks to corresponding agents based on subtask type, task dependency, resource status, and agent capabilities. The intelligent agent coordination module includes a predictive intelligent agent, an energy storage intelligent agent, a security intelligent agent, a carbon optimization intelligent agent, and a scheduling intelligent agent; The collaborative decision-making module is used to perform conflict detection, consistency verification, priority coordination, and result fusion on the output results of each agent. The scheduling output module is used to output the final scheduling scheme through a standardized service interface.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images