Enhancing HVAC energy management through multi-zone occupant-centric approach: A multi-agent deep reinforcement learning solution

Abstract

Occupant-centric HVAC control places a premium on factors including thermal comfort and electricity cost to guarantee occupant satisfaction. Traditional approaches, reliant on static models for occupant behaviors, fall short in capturing intra-day behavioral variations, resulting in imprecise thermal comfort evaluations and suboptimal HVAC energy management, especially in multi-zone systems with diverse occupant profiles. To address this issue, this paper proposes a novel occupant-centric multi-zone HVAC control approach that intelligently schedules cooling and heating setpoints using Multi-agent Deep Reinforcement Learning (MADRL). This approach systematically takes into account stochastic occupant behavior models, such as dynamic clothing insulation adjustments, metabolic rates, and occupancy patterns. Simulation results demonstrate the efficacy of the proposed approach. Comparative case studies show that the proposed MADRL-based, occupant-centric HVAC control reduces electricity costs by 51.09% compared to rule-based approaches and 4.34% compared to single-agent DRL while maintaining multi-zonal thermal comfort for occupants.