基于深度强化学习的电网紧急控制策略研究-中国电机工程学报.PDFVIP

第 38 卷 第 1 期 中 国 电 机 工 程 学 报 Vol.38 No.1 Jan. 5, 2018 2018 年 1 月 5 日 Proceedings of the CSEE ©2018 Chin.Soc.for Elec.Eng. 109 DOI ：10.13334/j.0258-8013.pcsee.171747 文章编号：0258-8013 (2018) 01-0109-11 中图分类号：TM 71 基于深度强化学习的电网紧急控制策略研究 刘威，张东霞，王新迎，侯金秀，刘丽平 ( 中国电力科学研究院有限公司，北京市 海淀区 100192) A Decision Making Strategy for Generating Unit Tripping Under Emergency Circumstances Based on Deep Reinforcement Learning LIU Wei, ZHANG Dongxia, WANG Xinying, HOU Jinxiu, LIU Liping (China Electric Power Research Institute, Haidian District, Beijing 100192, China) ABSTRACT: This paper proposed a kind of method for 0 引言 improving generating unit tripping strategy using deep 随着我国电网的快速发展，新能源渗透率不断 reinforcement learning. The method is data driven, only power gird environmental information is needed. Firstly, basic 提高，电网结构日趋复杂。传统的电网暂态稳定分 principle and framework of reinforcement learning was briefly 析控制技术与电网高速发展带来的新的稳定性问 introduced and Q-Learning was expounded in detail. Secondly, 题之间的矛盾日益凸显。文献[1]将风电不确定性归 the fundamental idea of deep learning was introduced. Then the 纳为风速的不确定性、风电转换中的不确定性和风 deep convolutional neural network was used to extract features 电系统外部的不确定性，对电网影响包括频率、电 for power system during transient process. Thirdly, the deep 压、暂态稳定性、电能质量、电力市场、充裕性、 reinforcement learning model was constructed by combining deep learning with reinforcement learning, and the double Q 节能减排等多方面影响。文献[2]介绍了新能源并网 model and dueling Q model was used to improve the 引发电网次同步振荡和超同步振荡，指出物理建模 performance about Q-Learnin