An emergency control strategy for undervoltage load shedding of power system: A graph deep reinforcement learning method

Undervoltage load shedding (UVLS) is the last line of defense to ensure safe and stable operation power system. The existing UVLS technique has difficulty adapting generalizing new topology variation scenarios network, which greatly affects reliability control strategy. This paper proposes a emergency scheme based on graph deep reinforcement learning method named GraphSAGE-D3QN (graph sample aggregate-double dueling q network). During offline training, GraphSAGE-based feature extraction mechanism grid with designed that can better capture changes in system state characteristics. A novel framework D3QN developed for modeling, reduces overestimations action leads effect. Then, online decision-making achieved. simulation results modified IEEE 39-bus 300-bus show proposed always provide more economical reliable strategies networks variations achieves benefits both adaptability generalization performances previously unseen scenarios.