This project will focus on a power service restoration problem in the design of the smart grids. Due to increasingly severe weather events and cyber-physical security threats, a more resilient and reliable power system is needed to ensure the continuous operation and availability of power applications and services. Traditionally, a lot of manual interruptions are required to restore large-scale power outages which may last hours or days. Therefore, we are interested in developing a self-healing smart grid that can automatically and rapidly restore problematic components. The focus will be on designing power service restoration plans for distribution systems with microgrids. Similar to other design problems in the smart grids, a service restoration plan needs to consider uncertainties from renewable power generation and load, and adopts the control action according to the uncertainty. Traditional computational tools may not be able to account for such complex and high uncertain environment. Here, we apply deep reinforcement learning to tackle this power service restoration problem. The goal is to derive a restoration plan (policy) for all the switchable components to maximize the total restored energy.
