Integrating Reinforcement Learning and Optimal Power Dispatch to Enhance Power Grid Resilience

Abstract

Power grids are vulnerable to extreme events that may cause the failure of multiple components and lead to severe power outages. It is of practical importance to design effective restoration strategies to enhance the power grid resilience. In this brief, we consider different time scales of various restoration methods and propose an integrated strategy to maximize the total amount of electricity supplied to the loads in the recovery process. The strategy properly combines the slow restoration method of component repair and the fast restoration method of optimal power dispatch. The Q-learning algorithm is used to generate the sequential order of repairing damaged components and update the network topology. Linear optimization is used to obtain the largest amount of power supply on given network topology. Simulation results show that our proposed method can coordinate the available resources and manpower to effectively restore the power grid after extreme events.