Optimal real-time Voltage/Var control for distribution network: Droop-control based multi-agent deep reinforcement learning
Peishuai Li, Jiawei Shen, Zaijun Wu, Minghui Yin, Yanhao Dong, Jing Han
Abstract
The photovoltaics (PVs) known as clean power generation has been highly penetrated in distribution networks, which motivates the network develops towards high uncertainty with complex real-time operation status. In this paper, an optimal real-time Voltage/Var control (ORT-VVC) method is proposed to reduce power losses and mitigate voltage fluctuations by optimizing the reactive power output of the PV inverter. And the ORT-VVC is implemented via a novelly proposed droop-control based multi-agent deep reinforcement learning (DC-MADRL). The VVC structure is constructed in distributed pattern with network partition and the multi-agents are established to rule the sub-networks via rare information communication between neighbors. Then, the PV control model is established as a Q-V droop control model with adjustable parameters. The MADRL is employed to optimize the PV controlling parameters instead of optimizing PV reactive power output directly, which aims to improve the real-time VVC performance for distribution network. The DC-MADRL model is established and solved via the multi-agent deep deterministic policy gradient (MADDPG) algorithm. Finally, numerical simulations are performed on the IEEE 123-bus test system to demonstrate the effectiveness of the proposed method.