深度强化学习在微电网系统调控中的应用综述  

Review on Applications of Deep Reinforcement Learning in Regulation of Microgrid Systems

作　　者：张有兵[1] 林一航 黄冠弘 杨晓东[1] 翁国庆[1] 周致言 ZHANG Youbing;LIN Yihang;HUANG Guanhong;YANG Xiaodong;WENG Guoqing;ZHOU Zhiyan(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,Zhejiang Province,China)

机构地区：[1]浙江工业大学信息工程学院,浙江省杭州市310023

出　　处：《电网技术》2023年第7期2774-2787,共14页Power System Technology

基　　金：国家自然科学基金项目(52007074)。

摘　　要：随着微电网系统的复杂性、随机性和数据维度不断增加,传统的模型驱动方法可能存在建模难度高、计算效率低、易受不确定因素干扰等缺点,因此难以对微电网进行精准高效的优化调控。近年来,深度强化学习迅猛发展,作为一种数据驱动的方法,因其结合了深度学习和强化学习的优势,可以在大规模场景和有限信息下,学习大量高维的和具有不确定性的数据来解决决策问题,克服传统模型驱动方法中存在的问题,实现对微电网的实时控制和反馈。该文首先对深度强化学习的基本思想、算法和优势进行了概述,然后从多组件协调运行、能源管理、能量交易、故障检测与恢复、发电功率预测和系统控制保护6个方面回顾总结了将深度强化学习应用于微电网系统调控中的现有研究成果,并与传统的求解方法进行了对比,分析总结了深度强化学习在微电网优化调控中的优势。最后,从提升硬件设施、更新优化算法、解决隐私问题、提高迁移能力4个方面提出展望。With the increasing complexity,randomness and data dimensions of microgrid systems,the traditional model-driven methods have shown disadvantages such as high modeling difficulty,low computational efficiency and big vulnerability to interferences of the uncertain factors,so it is difficult to optimize and regulate the microgrids accurately and efficiently.In recent years,the deep reinforcement learning has developed rapidly.As a data-driven method,it combines the advantages of deep learning and reinforcement learning.It can solve the decision-making problems by learning a large amount of high-dimensional and uncertain data.Besides,it can overcome the problems that exist in the traditional model-driven methods,and realize real-time control and feedback of microgrids.This paper first gives an overview of the basic ideas,the algorithms and the advantages of deep reinforcement learning.Then,it reviews and summarizes the existing research results of applying deep reinforcement learning to the microgrids in six aspects:the coordinated operation of multiple components,the energy management,the energy trading,the fault detection and recovery,the power generation forecast and the system control protection.It also compares with the traditional solution methods,and draws a conclusion of the superiorities of deep reinforcement learning in the optimization and regulation of microgrids.Finally,this paper prospects deep reinforcement learning from four aspects:the hardware facility construction,the algorithm update and optimization,the privacy improvement,and the migration capability.

关 键 词：深度强化学习 人工智能 微电网 优化调控 

分 类 号：TM721[电气工程—电力系统及自动化]