Swarm Reinforcement Learning for Operational Planning of Energy Plants for Small and Mid-Sized Building Energy Management Systems

Abstract

This paper proposes operation planning of energy plants by swarm reinforcement learning in order to realize successful BEMS for small and mid-sized buildings. It usually takes many man-hours to develop an evolutionary computation based program and develop a model considering facility characteristics and so on for an energy management system, while engineering man-hours can be reduced and appropriate operational planning can be expected to be realized by a versatile program of swarm reinforcement learning without consideration of facility characteristics and so on. Moreover, the results of the proposed methods are compared with those of a basic Q learning based method and a basic particle swarm optimization (PSO) based method. It is verified that energy cost can be more reduced by one of the proposed methods (PSO-Q based method) than those by the original Q-learning based method. Since the rates to the whole cost are large in case of small and mid-sized buildings, the proposed swarm reinforcement learning based methods can contribute to successful BEMS for small and mid-sized buildings.