强化学习理论在电力系统中的应用及展望  

Application and development of reinforcement learning theory in power systems

作　　者：余涛[1] 周斌[1] 甄卫国[1] 

机构地区：[1]华南理工大学电力学院,广东广州510640

出　　处：《电力系统保护与控制》2009年第14期122-128,共7页Power System Protection and Control

基　　金：国家自然科学基金项目(50807016);广东省自然科学基金博士启动基金项目(06300091)~~

摘　　要：强化学习理论是人工智能领域中机器学习方法的一个重要分支,也是马尔可夫决策过程的一类重要方法。所谓强化学习就是智能系统从环境到行为映射的学习,以使奖励信号(强化信号)函数值最大。强化学习理论及其应用研究近年来日益受到国际机器学习和智能控制学术界的重视。系统地介绍了强化学习的基本思想和算法,综述了目前强化学习在安全稳定控制、自动发电控制、电压无功控制及电力市场等方面应用研究的主要成果与方法,并探讨了该课题在电力系统运行控制中的巨大潜力,以及与经典控制、神经网络、模糊理论和多Agent系统等智能控制技术的相互结合问题,最后对强化学习在电力科学领域的应用前景作出了展望。Reinforcement Learning （RL） theory is an important branch of the machine learning in the field of artificial intelligence, which is also the general method to deal with Markov Decision Process problems. RL takes learning as trial and error process so as to maximize the reward value function by choosing an action depending on the state. In recent years, RL and its application are received increasing attention of international academia. In order to propel the further study on the aspect of RL in power systems, this paper introduces the basic idea and algorithms systematically, the main achievements of RL are surveyed in security and stability control, automatic generation control voltage and reactive power control and electricity market respectively. Furthermore, the paper discusses the application potentials of RL in power system operation and control, and the combination of RL with classical control, ANN, fuzzy theory and multi-agent system. Meanwhile, the prospect of RL theory in power system is brought forward.

关 键 词：人工智能 强化学习 马尔可夫决策过程 随机最优控制 电力系统 

分 类 号：TM7[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]