电力市场智能模拟中代理决策模块的实现  

Realization of Decision-making Module in Agent-based Simulation of Power Markets

作　　者：陈皓勇[1] 杨彦[1] 张尧[1] 王野平[1] 荆朝霞[1] 陈青松[1] 

机构地区：[1]华南理工大学电力学院,广东省广州市510640

出　　处：《电力系统自动化》2008年第20期22-26,共5页Automation of Electric Power Systems

基　　金：国家重点基础研究发展计划(973计划)资助项目(2004CB217905);国家社会科学基金资助项目(04CJ2012)~~

摘　　要：在日前交易方式下,发电厂商为了追求长期最大利润,竞价策略显得尤其重要。通常,发电厂商运用的策略过于复杂,难以用传统的博弈论方法来建模。人工智能中强化学习Q-learning算法是一种自适应的学习方法,使代理能够通过不断与环境进行交互所得到的经验进行学习,适合在电力市场智能模拟中运用。文中在开放源代码的电力市场智能模拟平台AMES上,增加了发电厂商代理基于Q-learning的竞价决策程序模块,并在5节点测试系统上进行模拟。实验结果表明,运用基于Q-learning算法竞价决策使代理可以较好地模拟发电厂商的经济特性,且在相同条件下表现出比AMES原有的VRElearning算法更强的探索能力。One of the mast important issues of power suppliers is to find bidding strategy in day-ahead electricity auction market to maximize their profit. Usually, the strategies used by power suppliers are too complex to model them by standard techniques based on game theory. Q-learning algorithm can find the optimal strategy through agents＇ experience, which is obtained from interaction of environment directly. This property makes this algorithm well suitable for dealing with the decision-making problems of power suppliers. AMES, a famous open-source experimental platform, which gives researchers full access to implementation. In this work, a program module, which helps the agents make decision through Q-learning algorithm, has been added into the AMES. A five-bus power system is used for case studies. The results show that the economic characteristics of power suppliers can be simulated by the actions of agents who use Q-learning algorithm. Furthermore, under the same conditions, the Q-learning algorithm illustrates the better ability of exploration than VRE learning algorithm.

关 键 词：智能代理模拟 竞价策略 电力拍卖市场 Q—learning算法 VRE learning算法 

分 类 号：TM743[电气工程—电力系统及自动化]