Dynamic reconfiguration of shipboard power systems using reinforcement learning

Abstract

A novel approach for the automatic reconfiguration of shipboard power systems (SPS) based on Q-learning has been investigated. Using this approach it is possible to obtain an optimal set of switches to open/close, in order to restore power to the loads, such that the weighted sum of the power delivered to the loads is maximized. This approach differs significantly from other methods previously studied for reconfiguration as it is a dynamic technique that produces not only the final reconfiguration, but also the correct order in which the switches are to be changed. Simulation results clearly demonstrate the effectiveness of this method.