发电商的模仿学习竞价策略  

Bidding Strategy for Generating Firm Based on Imitation Learning

作　　者：曾亮[1] 齐翔[1] 陈迎春[1] 

机构地区：[1]华中科技大学控制科学与工程系,湖北省武汉市430074

出　　处：《中国电机工程学报》2008年第31期74-80,共7页Proceedings of the CSEE

基　　金：国家自然科学基金资助项目(60574088;60774036)。~~

摘　　要：在重复博弈的电力市场中,发电商的竞价策略和个体学习行为是个复杂的动态问题。分析发电商竞价策略这一热点问题的最新研究思路和仿真方法。为模拟发电商的自主学习和市场的动态演化过程,引入博弈学习理论,建立基于模仿学习的发电商竞价策略模型,分析不同信息条件下的学习动态过程,及重复竞标时发电商模仿学习对博弈演变及市场均衡的影响。仿真结果表明模仿学习具有较好的收敛性。另外,完全信息条件下,发电商全局学习能引导市场趋于瓦尔拉斯均衡和古诺–纳什均衡二者之间的一个均衡状态;非完全信息条件下,发电商全局学习的结果是市场总体行为收敛到古诺–纳什均衡,而发电商的个体行为则与古诺-纳什均衡状态存在偏差。In electricity markets with repeated games, GenCos＇ bidding strategies and learning behaviors are complex dynamic problems. The research thinking and simulation methods of the above hot issues are reviewed. To simulate and study GenCos＇ learning behaviors and the dynamic evolution process of electricity markets, a GenCo＇ bidding model based on imitation learning is presented. By importing the idea of learning in games and applying the proposed bidding model, the dynamic learning process, and its influence on the evolution process of games and market equilibrium in repeated bidding are analyzed under various information conditions. Simulation results show the convergency and efficiency of imitation learning. Meanwhile, GenCos＇ imitation learning in global view will produce quite different outcomes under full and partial information conditions. Under full information condition, the general market behavior converges at an intermediate equilibrium state between Walrasian equilibrium and Cournot-Nash equilibrium; under partial information condition, the general market behavior converges at Cournot- Nash equilibrium while the individual behavior of a GenCo diverges.

关 键 词：博弈学习理论 模仿学习 竞价策略 博弈演变 市场均衡 古诺-纳什均衡 

分 类 号：TM73[电气工程—电力系统及自动化]