Optimal Scheduling of Battery Energy Storage Systems Using a Reinforcement Learning-based Approach

Abstract

This article proposes a novel energy management algorithm that controls the battery energy storage system (BESS) and on-grid supply. It employs the deep-Q-network agent with prioritized experience replay, and its efficacy is validated and verified by comparison to a benchmark method for mixed integer linear programming. The grid and energy storage systems are governed by switching operations initiated by BESS controllers via the automatic transfer switch. The primary objective is to accomplish optimal scheduling of batteries one day in advance to reduce electricity costs while maintaining battery health and primary power supply reliability. The methods proposed in this work provide practicable grid and battery operation patterns that test all conceivable planning scenarios for energy storage operation. Finally, a comparative analysis is performed to evaluate the efficacy of the proposed BESS operation scheduling methods.