JAUE2022-096: Using Deep Reinforcement Learning to Optimize the Economic Performance of Residential Microgrid Energy System

Abstract

This paper presents a reinforcement learning approach for optimizing operational performance of a real zero energy house considering minimizing energy cost and increasing the local photovoltaic(PV) self-consumption. This paper formulates a DQN algorithm for PV-battery real-time control, the management problem is formulated as Markov chain process, detail states, actions and rewards are designed properly. In the optimization of energy operation, the deep reinforcement learning algorithms show an advantage over solving continuous sequential decision problems. The results show that the photovoltaic consumption rate has been improved. This increase was achieved without causing any perceived discomfort to the occupants. Besides, through managing the charge and discharge power flows of the battery, the energy cost of consumer was reduced by 6.7%. this work verified the effectiveness of proposed algorithm in building energy system control in terms of energy saving and cost reduction