Adaptive power system scheduling using transfer reinforcement learning in generator maintenance scenarios

To address the limitations in generalization of existing model-free power system scheduling methods, which struggle to meet the heterogeneous operational demands such as generator maintenance-induced power changes, this paper proposes an optimized power system scheduling method based on transfer reinforcement learning. First, a deep reinforcement learning-based power system scheduling model is constructed, taking into account the operational characteristics, optimization objectives, and security constraints of power system scheduling. Next, a deep deterministic policy gradient (DDPG) algorithm is developed for training the scheduling model. Finally, considering the operational requirements of generator maintenance, an adaptive scheduling method based on flexible scaling network reconfiguration in transfer reinforcement learning is designed. The effectiveness of the proposed method is validated through simulations on a 126-bus system.