Research on Emergency Control Strategy for Transient Instability of Power System Based on Deep Reinforcement Learning and Graph Attention Network

Emergency control plays a vital role in preserving the transient stability of power systems in the aftermath of severe faults. Deep reinforcement learning (DRL) emerges as a potent tool for addressing online emergency control in power systems. However, existing DRL-based emergency control methods have not fully exploited the potential of the grid topology during system operation, despite its significant impact on transient stability. To address this gap, this paper proposes a novel approach that combines graph attention networks with deep reinforcement learning. By incorporating the graph attention mechanism into the neural networks of DRL, we aim to effectively model the network topology of power systems. Furthermore, we design corresponding action space, state space, and reward function tailored to establish a superior strategy for mitigating transient instability in power systems. Finally, we validate the efficacy of the proposed approach through comparative analysis with traditional DRL methods using the IEEE 39-bus system as a testbed.