Reinforcement Learning for Multiagent-based Residential Energy Management System

Abstract

In Smart Grid (SG), energy management in residential houses has gained widespread popularity with the increasing population and demand for energy. Residential Energy Management System (REMS) can lead to higher efficiency and lower operating costs for the multi-carrier energy supply (i.e., gas and electricity). Several approaches exist for single carrier REMS, however, it is quite challenging to develop a multi-carrier REMS due to the dynamic consumption environment. So, this paper proposes a MultiAgent-based Residential Energy Management scheme, i.e., MA-REM for a multi-carrier energy system. It benefits each energy component, i.e., gas and electricity with different Demand Response Program (DRP), which accelerates by employing Reinforcement Learning (RL) methodology. The proposed MA-REM scheme uses Q-learning based on a dynamic pricing mechanism for optimal energy management in REMS to reduced energy cost (14.97%) and energy consumption. The effectiveness of the proposed MA-REM scheme is evaluated by comparing it with existing approaches in terms of energy consumption and energy cost for electricity & gas.