HVACLearn

Abstract

In this paper, we present a Reinforcement Learning (RL) based Occupant-Centric Controller (OCC) for thermostats, HVACLearn. Monitoring indoor air temperature, occupancy, and thermal vote, the agent learns the unique occupant behavior and indoor environments and calculates adaptive thermostat set-points to balance between occupant comfort and energy efficiency. We simulated HVACLearn performance in a single occupant office with occupant behavior models from the literature (i.e., occupancy and thermal vote). Compared to a reference controller, HVACLearn reduced the number of button presses (too hot) significantly, while consuming same or less cooling energy. For the heating, HVACLearn resulted in almost same number of button presses (too cold) with slightly less heating energy consumption.