基于已知电网场景分段拟合智能体智能评估与自主进化方法  

Agent’s intelligence assessment and self-evolution method based on known power grid scenarios piecewise fitting

作　　者：张天韵 张俊[1] 许沛东 高天露 ZHANG Tianyun;ZHANG Jun;XU Peidong;GAO Tianlu(School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学电气与自动化学院,湖北武汉430072

出　　处：《电力自动化设备》2025年第4期177-185,共9页Electric Power Automation Equipment

基　　金：国家电网公司科技项目(5700-202217190A-1-1-ZN)。

摘　　要：针对智能体的智能评估与自主进化过程中较难满足海量训练测试场景的需求这一实际工程问题,提出了已知场景下的电网调控智能体的智能评估与自主进化技术方法。该方法利用智能量化评估与自主进化系统在已知场景中进化生成最优电网调控智能体,依据智能体智能函数向目标函数分段拟合逼近原理,抽取该最优智能体适用的训练测试场景集,以获取已知场景下电网调控智能体的评估、进化和各批次电网场景的最优智能体,从而满足电网调控智能体的实际工程应用。算例结果表明,利用该方案可以确保在已知场景条件下,通过智能评估与自主进化,获得具有较高智能水平和较强电网调控能力的智能体。所提方案获得的智能体-场景集,其校正控制智能体在对应已知场景中的智能评估结果均在80分以上。As for the practical engineering problem in difficultly meeting the needs of massive training and testing scenarios in the process of intelligence assessment and self-evolution of agents,the technical method of intelligence assessment and self-evolution for power grid control agents in known scenarios is proposed.This method uses the parallel system based quantitative assessment and self-evolution(PLASE)system to evolve and generate the best power grid regulation control agents in the known scenarios,accor-ding to the principle of piecewise fitting approximation of the agent’s intelligence function to the objective function,extracts the applicable training and testing scenarios to the best agent.In this way,the intelligence assessment and self-evolution of power grid control agents and the best agent for each batch of power grid scenarios could be obtained,so as to meet the practical engineering application of agents for power grid control.The results show that using this method can ensure obtaining the agents with high intelligence level and strong power grid control ability through the intelligence assessment and self-evolution in known scenarios.In the agent-scenario set gained by the proposed method,the scores of intelligence assessment results of the regulation control agent in the corresponding known scenarios are all above 80.

关 键 词：电网调控 已知场景 智能量化评估 基于智能评估的自主进化 电网调控智能体 

分 类 号：TM76[电气工程—电力系统及自动化]