Reinforcement Learning Based Reactive Power Real-Time Dispatch Optimization in Distribution Networks

The access of large-scale distributed generation (DG) makes the reactive power optimal control strategy of distribution networks different from traditional distribution networks. Aiming at the problem of real-time reactive power dispatch for short-time scale scheduling, this paper proposes a Volt-Var control (VVC) method of distribution networks based on multi-agent deep reinforcement learning (MADRL). Firstly, the decentralized partially observable Markov decision process (Dec-POMDP) is designed based on the VVC model, and then the multi-agent twin delayed deep deterministic policy gradient (MATD3) algorithm is used to solve the Dec-POMDP model. The proposed method can intelligently generate the real-time VVC dispatch strategy according to the random change of generation and load. It has good real-time performance and control economy. Finally, the effectiveness of the proposed method is verified by simulation examples.