Deep Reinforcement Learning for Volt/VAR Control in Distribution Systems: A Review

Abstract

An increasing number of distributed generators are integrated to distribution systems, which has a huge impact on the network power flow and leads to severe voltage fluctuation problems. Volt/VAR control is regarded as an effective approach to improve voltage quality and reduce power loss. Deep reinforcement learning is a data-driven approach to effectively solve decision-making problems. The application of deep reinforcement learning in volt/VAR control circumvent the accurate knowledge of network information, and is endowed with less computational burden. This paper provides a general review of the application of deep reinforcement learning in volt/VAR control in terms of basic notations, Markov decision process formulations, and control framework. Future directions are also discussed.