基于强化学习的综合能源系统智能体设计  

Design of intelligent agents for integrated energy systems based on reinforcement learning

作　　者：张磊 张继权 李一明 徐英[2] 刘秉祺 ZHANG Lei;ZHANG Jiquan;LI Yiming;XU Ying;LIU Bingqi(State Grid Jilin Electric Power Co.,Ltd.,Changchun 130021,China;School of Electrical Engineering and Automation,Harbin Institute of Technology,Harbin 150001,China;Beijing QU Creative Technology Co.,Ltd.,Beijing 100025,China)

机构地区：[1]国网吉林省电力有限公司,吉林长春130021 [2]哈尔滨工业大学电气工程及自动化学院,黑龙江哈尔滨150001 [3]北京清大科越股份有限公司,北京100025

出　　处：《电子设计工程》2024年第12期145-149,共5页Electronic Design Engineering

基　　金：国家电网公司科技项目(52230021001800)。

摘　　要：为满足综合能源系统现货市场的交易需求,文中构建了申报策略强化学习的智能体。该智能体可以完成市场环境与申报策略间的动态匹配,进而灵活选取最适宜的申报策略。同时还根据综合能源系统申报策略的制定需求,设计了智能体的环境变量、动作空间和奖励函数。基于某省电网实际数据构造了智能体的仿真数据集,并从主流技术路径下筛选了8种申报方法构建智能体的动作策略库来进行实验对比。仿真结果表明,动态申报智能体策略的理想度相比其他申报方法高6%以上,且该方法在策略上的准确性良好,决策时间也能够满足现货市场的决策要求。In order to meet the transaction demand of the spot market of the integrated energy system,an agent for strengthening the learning of declaration strategy is constructed in this paper.This agent can dynamically match the market environment with the declaration strategy and flexibly select the most appropriate declaration strategy.At the same time,the environment variables,action space and reward function of the agent are designed according to the formulation requirements of the declaration strategy of the integrated energy system.Based on the actual data of a provincial power grid,an agent simulation dataset is constructed,and eight declaration methods are screened from the mainstream technology path to build an agent action strategy library.The simulation results show that the ideality of the dynamic declaration agent strategy is more than 6%higher than other declaration methods,the accuracy of the strategy is good,and the decision⁃making time can meet the decision⁃making requirements of the spot market.

关 键 词：智能体 现货市场交易 申报策略 强化学习 综合能源 

分 类 号：TN929.5[电子电信—通信与信息系统] TP311[电子电信—信息与通信工程]