| 引用本文: | 符杨,郭笑岩,米阳,李振坤,袁明瀚.基于强化学习的直流微电网分布式经济下垂控制[J].电力自动化设备,2021,41(11): |
| FU Yang,GUO Xiaoyan,MI Yang,LI Zhenkun,YUAN Minghan.Distributed economic droop control for DC microgrid based on reinforcement learning[J].Electric Power Automation Equipment,2021,41(11): |
|
| 摘要: |
| 针对直流微电网在传统下垂控制下整体运行成本偏高且存在电压静态偏差的问题，提出一种基于强化学习的完全分布式经济下垂控制策略。通过分布式一致性算法寻找系统最优经济运行点，并利用偏差调整和Adam算法优化经济下垂系数，以提高计算的速度与效率；通过改进强化学习原理进行电压的二次优化控制。利用MATLAB/Simulink搭建具有不同运行成本单元的直流微电网仿真模型，结果验证了所提策略的有效性和优越性以及即插即用特性。 |
| 关键词: 直流微电网 经济下垂控制 一致性算法 完全分布式 强化学习 |
| DOI：10.16081/j.epae.202109024 |
| 分类号:TM73;TM727 |
| 基金项目:国家自然科学基金面上项目(61873159)；上海市科委项目(18020500700)；上海绿色能源并网工程技术研究中心项目(13DZ2251900);“电气工程”上海市Ⅱ类高原学科项目；上海市电站自动化技术重点实验室项目 |
|
| Distributed economic droop control for DC microgrid based on reinforcement learning |
|
FU Yang1, GUO Xiaoyan2, MI Yang1, LI Zhenkun1, YUAN Minghan3
|
|
1.College of Electric Power Engineering, Shanghai University of Electric Power, Shanghai 200090, China;2.State Grid Weihai Electric Power Company, Weihai 264200, China;3.School of Mechanical and Electrical Engineering and Automation, Shanghai University, Shanghai 200072, China
|
| Abstract: |
| Aiming at the problems of high overall operation cost and static voltage deviation of DC microgrid under the traditional droop control, a fully distributed economic droop control strategy based on reinforcement learning is proposed. The optimal economic operation point of the system is searched by the distributed consensus algorithm, and the economic droop coefficient is optimized by the deviation adjustment term and Adam algorithm to improve the calculation speed and efficiency. The secondary optimal control of voltage is carried out by improving the reinforcement learning principle. The simulation model of DC microgrid with units of different operation costs is built by MATLAB/Simulink, and the results verify the effectiveness and superiority of the proposed strategy, and its plug-and-play characteristics. |
| Key words: DC microgrid economic droop control consensus algorithm fully distributed reinforcement learning |
