A Power System Corrective Control Method Based on Evolutionary Reinforcement Learning

Corrective control becomes more and more important for power systems due to the increasing penetration of renewable energy. As an effective method in control problems, the Reinforcement learning method is considered to provide decisions for corrective control in power networks. However, the large size of action and state space, as well as the sparse reward problem in corrective control limits the application of the RL method. This paper proposed an evolutionary reinforcement learning method. Combining the evolutionary algorithm and Reinforcement learning methods, this method decreases the training difficulty of reinforcement learning. The experiments based on the Grid2op environment show that the proposed method has a longer operation time than other traditional methods in the corrective control of power networks.