The Potential of Control Models Based on Reinforcement Learning in the Operating of Solar Thermal Cooling Systems

Abstract

The objective of this research work was to investigate the potential of control models based on reinforcement learning in the optimization of solar thermal cooling systems (STCS) operation through a case study. In this, the performance of the installation working with a traditional predictive control approach and with a reinforcement learning (RL)-based control approach was analyzed and compared using a specific realistic simulation tool. In order to achieve the proposed objective, a control system module based on the reinforcement learning approach with the capacity for interacting with the aforementioned realistic simulation tool was developed in Python. For the studied period and the STCS operating with a control system based on RL, the following was observed: a 35% reduction in consumption of auxiliary energy, a 17% reduction in the electrical consumption of the pump that feeds the absorption machine and more precise control in the generation of cooling energy regarding the installation working under a predictive control approach. Through the obtained results, the advantages and potential of control models based on RL for the controlling and regulation of solar thermal cooling systems were verified.