Data-Driven Multi-Agent Deep Reinforcement Learning for Distribution System Decentralized Voltage Control With High Penetration of PVs

This paper proposes a novel model-free/data-driven centralized training and decentralized execution multi-agent deep reinforcement learning (MADRL) framework for distribution system voltage control with high penetration of PVs. The proposed MADRL can coordinate both the real reactive power PVs existing static var compensators battery storage systems. Unlike DRL-based methods, our method does not rely on model during stages. is achieved by developing new interaction scheme between surrogate modeling original soft actor critic (MASAC) algorithm. In particular, sparse pseudo-Gaussian process few-shots measurements utilized to construct environment, i.e., flow model. data-driven no parameters are needed. Furthermore, MASAC enabled allows achieve better scalability dividing into different regions aid sensitivities voltage, where each region treated as an agent. also serves foundation execution, thus significantly reducing communication requirements only local required control. Comparative results other alternatives IEEE 123-nodes 342-nodes systems demonstrate superiority method.