摘要
微网在分布式新能源消纳、负荷优化、提高能源利用效率等方面具有重要作用。但新能源出力的间歇性、负荷侧用电行为的随机性导致微网成为一个动态的复杂系统,难以通过准确的物理模型刻画,给微网优化运行带来巨大挑战。深度强化学习(deep reinforcement learning,DRL)通过与环境交互试错寻找最优策略,不依赖于新能源出力和负荷的精确建模,适用于解决序贯决策问题,在求解含有大量不确定性的微网优化运行难题时具有优势。为此,从DRL原理、DRL在单个微网以及微网群优化运行中的应用进行了综述与分析,最后对应用中所面临的算法可解释性、奖励函数设置、用户隐私性等方面进行了展望。
Microgrid is of great significance for the local consumption of new energy sources such as wind and solar,optimization of load levels,and improvement of energy utilization efficiency.However,the coupling of multiple energy systems,the intermittency in the output of new energy power,the randomness of load-side power demand and behavior have resulted in the microgrid becoming a dynamic and complex system,which has brought challenge to the optimization of the microgrid.Deep reinforcement learning(DRL)finds the optimal strategy through trial and error interaction with the environment,which can avoid accurate modeling of uncertainty,it is also suitable for solving sequential decision problems,therefore,it has advantages in solving dynamic microgrid optimization operation problems with a large number of uncertainties.This paper mainly reviews and analyzes the application of deep reinforcement learning in microgrid optimization operation.Finally,it discusses interpretability,reward function settings,user privacy,transferability,combining model and model-free algorithms,multi-objective function weights which faced by the application of reinforcement learning in the optimization operation of microgrid.
作者
周翔
王继业
陈盛
王新迎
ZHOU Xiang;WANG Jiye;CHEN Sheng;WANG Xinying(China Electric Power Research Institute Co.,Ltd.,Haidian District,Beijing 100192,China;State Grid Digital Technology Holding Co.,Ltd.,Xicheng District,Beijing 100053,China)
出处
《全球能源互联网》
CSCD
2023年第3期240-257,共18页
Journal of Global Energy Interconnection
基金
国家电网有限公司总部科技项目(基于边云协同的微网群优化运行智能技术研究及应用,5700-202130263A-0-0-00)。
关键词
微网
深度强化学习
复杂系统
优化运行
microgrid
deep reinforcement learning
complex system
optimal operation
