Fundamental Study of Energy Management System using Deep Reinforcement Learning

Abstract

Deep reinforcement learning (DRL) is expected to be a promising optimal control method for multienergy management systems. However, no control method has been established for operation in a real environment. In addition, conventional DRL models may cause excessive charging and discharging, which may have a significant impact on actual equipment. Therefore, it is necessary to create models that are considered the actual equipment and faithfully reproduce the constraints of the actual equipment. In this study, home energy management system (HEMS) was constructed using a storage battery control model that is considered the actual equipment based on DRL. Simulation results confirmed that an actual storage battery is appropriately controlled by the proximal policy optimization (PPO) agent to the extent that they can be charged and discharged every 10 minutes.