Deep Reinforcement Learning-Based Adaptive Voltage Control of Active Distribution Networks with Multi-terminal Soft Open Point

• A data-driven voltage control method is proposed for M−SOPs based on DDPG. • A multi-dimensional action masking approach is designed for action space of M−SOPs. • A DDPG agent is built to mitigate the voltage violation caused by DG fluctuations. The integration of highly penetrated distributed generators (DGs) aggravates the rise of voltage violations in distribution networks. Connected by multi-terminal soft open points (M−SOPs), distribution networks gradually evolve into an interconnected flexible architecture with high controllability. Distribution networks with M−SOPs can exchange active power flexibly, and M−SOPs can provide local reactive power support to alleviate voltage violations. However, conventional model-based M−SOP optimization methods cannot regulate voltage profiles adaptively owing to the rapid fluctuations of DGs. In this paper, a data-driven voltage control method is proposed for M−SOPs using a deep deterministic policy gradient network (DDPG). First, the data-driven voltage control framework is proposed for M−SOPs based on DDPG. The M−SOP−based voltage control problem is reformatted as a Markov decision process (MDP) to construct the DDPG agent. Based on real-time measurement, the DDPG agent can adaptively regulate the M−SOP operation to address the frequent DG fluctuations. Then, a multi-dimensional and dynamic boundary action masking approach is proposed to address the complex coupling in the action space of M−SOPs. Finally, the effectiveness of the proposed method was verified using the IEEE 33-node system. The results show that the proposed method can adaptively alleviate the voltage fluctuations caused by rapid DG power variations.