Dynamic Load Shedding Strategy Using Distributional Deep Reinforcement Learning in Power System Emergency Control

Abstract

With the uncertainty and complexity of power system control improved, emergency control strategies are facing significant challenge on adaptiveness and robustness. This paper applies a distributional deep reinforcement learning method in dynamic load shedding, which allow agents at different buses take collaborative actions in a distributed way. These agents are centrally trained and separately executed, which can have mutual collaboration with others. To validate the effectiveness of DDRL, our simulations are implemented on an open-source platform named Reinforcement Learning for Grid Control. Furthermore, we make comparisons and analysis in the IEEE 39-bus system to evaluate the performance of distributional deep reinforcement learning, and the results have demonstrated that the proposed method have satisfied adaptiveness and robustness.