Data-driven voltage/Var control of active distribution networks: A comprehensive review of deep reinforcement learning methods

Voltage/var control (VVC) aims to regulate bus voltage and reduce energy loss in active distribution networks. However, conventional VVC is typically realized by model-based optimization methods, which heavily relies on the accurate system models and parameters. To address the bottlenecks and limitations of model-based methods, data-driven VVC has attracted much attention in recent years. This paper conducts a comprehensive review on deep reinforcement learning (DRL)-based data-driven VVC approaches. Firstly, the problem descriptions and mathematical models of VVC are explained. Secondly, the principles and characteristics of various VVC devices/resources are reviewed and categorized to two groups: tap changer based and power flow based. Thirdly, the DRL-based data-driven VVC solutions are systematically classified and comprehensively reviewed. Moreover, recent advancements of DRL in safety, performance and robustness enhancements are discussed. Finally, future research problems are suggested. • This paper provides the first systematic review of DRL methods for VVC in active distribution networks. • This paper formulates the VVC problem and reviews mechanisms of various VVC devices and resources. • This paper presents a taxonomy based on VVC devices, DRL algorithms, and communication structures. • This paper reviews recent DRL advances in safety, performance, and robustness, and outlines future research.