Thermal comfort management leveraging deep reinforcement learning and human-in-the-loop

Abstract

The design and implementation of effective systems devoted to the thermal comfort management in a building is a challenging task because they require to consider both objective and subjective parameters, tied for instance to human profile and behavior. This paper presents a novel approach for the management of thermal comfort in buildings by leveraging cognitive technologies, namely the Deep Reinforcement Learning paradigm. The approach is able to learn how to automatically control the HVAC system and improve people’s comfort. The learning process is driven by a reward that includes and combines an environmental reward, related to objective environmental parameters, with a human reward, related to subjective human perceptions that are implicitly inferred by the way people interact with the HVAC system. Simulation results aim to assess the impact of the two types of reward on the achieved comfort level.