GRIDDPG: A Graph-Reinforced Intelligent Deep Policy Gradient Approach for Transferable Power Grid Dispatch

This paper proposes a novel deep reinforcement learning approach, named GRIDDPG (Graph-Reinforced Intelligent Dispatch with Deep Policy Gradient), for intra-day optimization dispatch in power grids. The proposed method embeds a double-layer intersecting Graph Neural Network (GNN) within the Deep Deterministic Policy Gradient (DDPG) framework, enabling the learning of power grid topology and operating state representations that inform the dispatch decision-making process. Comprehensive experiments are conducted on the SG-126 power grid simulator under three representative scenarios to evaluate the performance and generalization capability of GRIDDPG. The results demonstrate that GRIDDPG consistently outperforms the standalone DDPG algorithm in terms of grid operation security, operating cost reduction, renewable energy accommodation, and overall reward while ensuring the stability of the power grid. Furthermore, the study explores the potential of transfer learning between scenarios and finds that GRIDDPG exhibits a promising generalization capability.