An Imitation Learning and Reinforcement Learning Approach to Power System Stabilizer

With the increasing complexity of power system structures, ensuring stability in power systems has become a significant challenge. This paper proposes a combined approach of imitation learning and reinforcement learning for designing power system stabilizers. The method involves collecting input-output data from conventional power system stabilizers as expert demonstrations and utilizing supervised learning to pretrain the policy, thereby initializing the reinforcement learning agent. The agent then interacts with the power system simulation environment to learn the optimal action policy and enhance power system stability. Simulation results on a four-machine two-area power system demonstrate that this technique outperforms traditional power system stabilizers in terms of performance.