不完全信息下基于多代理深度确定策略梯度算法的发电商竞价策略  

Bidding Strategy of Generation Companies Based on Multi-agent Deep Deterministic Policy Gradient Algorithm Under Incomplete Information

作　　者：员江洋 杨明[1] 刘宁宁[1] 张长行 黄诗颖 朱青 YUN Jiangyang;YANG Ming;LIU Ningning;ZHANG Changhang;HUANG Shiying;ZHU Qing(School of Electrical Engineering,Shandong University,Jinan 250012,Shandong Province,China;Shandong Institute of Innovation and Development,Jinan 250101,Shandong Province,China)

机构地区：[1]山东大学电气工程学院,山东省济南市250012 [2]山东省创新发展研究院,山东省济南市市250101

出　　处：《电网技术》2022年第12期4832-4842,共11页Power System Technology

基　　金：国家重点研发计划项目“基于多元柔性挖掘的主动配电网协同运行关键技术与仿真平台研究”(2019YFE0118400)。

摘　　要：在电力现货市场中,发电商竞价行为受多种因素综合影响,且由于信息受限无法做出最优决策,难以实现自身收益最大化。将发电商竞价决策行为建模为马尔科夫博弈过程,提出了基于多代理模型的发电商日前市场竞价模型,应用多代理深度确定性策略梯度(multi-agent deep deterministic policy gradient, MADDPG)算法,分别在IEEE-3节点算例和IEEE-30节点算例模拟发电侧竞价行为。算例分析表明,所提模型通信开销低、训练结果良好,可以在不完全信息条件下提高发电商收益,并实现发电侧报价的激励相容。In electricity spot market,many factors may affect the bidding behavior of generation companies(GENCOs).However,it is usually hard for the GENCOs to make the optimal decision in order to maximize their private profits due to the incomplete market information.In this paper,the bidding behavior of the GENCOs is transformed into a Markov game process and a day-ahead market bidding model of the GENCOs based on the multi-agent model is proposed.Furthermore,the multi-agent deep deterministic policy gradient(MADDPG) algorithm is employed to simulate the bidding behavior of the GENCOs in the IEEE 3-bus system and the IEEE 30-bus system.The example analysis shows that the proposed model,with its lower communication costs and better training effects,can improve the profits of the GENCOs with incomplete information and realize the incentive compatibility of bidding at the generation side.

关 键 词：电力现货市场 多代理模型 不完全信息 多代理深度确定性策略梯度 最优报价策略 激励相容 

分 类 号：TM73[电气工程—电力系统及自动化]