LAUSR

Abstract Commercial refrigeration systems consume a substantial amount of electrical energy, resulting in high indirect global warming impact due to greenhouse gases emissions. The multitude of different system configurations, system complexity, component wear, and changing operating conditions make efficient operation of this kind of refrigeration systems a difficult task. This paper presents an investigation of machine learning for supervisory control of a supermarket refrigeration system. In particular, a reinforcement learning algorithm for a CO2 booster refrigeration system is designed by exploiting the Matlab-based “SRSim” simulation tool. The reinforcement learning controller learns to operate the refrigeration system based on the interaction with the environment. The analysis shows that learning control is a feasible model-free technique to find a suitable control strategy for demand-side management in a smart grid scenario.