Cooperative Multi-Agent Deep Reinforcement Learning for Adaptive Decentralized Emergency Voltage Control

Under voltage load shedding (UVLS) for power grid emergency control builds the last defensive perimeter to prevent cascade outages and blackouts in case of contingencies. This letter proposes a novel cooperative multi-agent deep reinforcement learning (MADRL)-based UVLS algorithm in an adaptive decentralized way. With well-designed input signals reflecting the voltage deviation, newly structured neural networks are developed as intelligent agents to obtain control actions and their probabilities to accommodate high uncertainties in volatile power system operations. Moreover, the interaction among the agents for coordinated control is implemented and refined by a state-of-the-art attention mechanism, which helps agents concentratively learn effective interacted information. The proposed method realizes decentralized coordinated control, adapting to extremely high uncertainties. Case studies on an IEEE benchmark system indicate the superior performance of the proposed algorithm.