Deep-Reinforcement-Learning-Based Two-Timescale Voltage Control for Distribution Systems

Because of the high penetration renewable energies and installation new control devices, modern distribution networks are faced with voltage regulation challenges. Recently, rapid development artificial intelligence technology has introduced solutions for optimal problems dimensions dynamics. In this paper, a deep reinforcement learning method is proposed to solve two-timescale problem. All variables assigned different agents, discrete solved by Q network (DQN) agent while continuous deterministic policy gradient (DDPG) agent. agents trained simultaneously specially designed reward aiming at minimizing long-term average deviation. Case study executed on modified IEEE-123 bus system, results demonstrate that algorithm similar or even better performance than model-based scheme computational efficiency competitive potential online application.