Hierarchical Reinforcement Learning Approach for Real-Time Optimal Power Flow

With the development of artificial intelligence, deep reinforcement learning (DRL) has shown a promising prospect for solving the real-time dispatch problem in real-world power systems. However, the decisions generated by the traditional DRL method still suffer from challenges of insufficient robustness and poor exploration efficiency. In order to cope with the above issues, this paper proposes a physics-informed neural network (PINN)-based DRL method through injecting physical knowledge into the loss function of the action network, which improves its robustness and guides its exploration. The effectiveness of the proposed algorithm is verified through numerical simulations conducted on the IEEE 39-bus system.