Adaptive Indoor Environment Regulation: A Reinforcement Learning Approach

This paper presents a feasibility‐aware reinforcement learning framework for multizone HVAC regulation, designed to jointly optimize thermal conditions, indoor air quality, and energy use under explicit operational constraints. Building‐derived limits, including airflow capacities, thermal load bounds, and equipment restrictions, are embedded directly into the learning environment so that control policies are trained within an admissible operating space. The framework further accounts for cross‐zone interactions and shared HVAC resources, enabling coordinated decision‐making across multiple thermal zones. The proposed approach outperformed a rule‐based baseline by reducing energy demand and improving pollutant control while maintaining acceptable indoor conditions during occupied periods. The study shows that reinforcement learning becomes more credible for intelligent building control when physical feasibility and multizone coordination are treated as integral parts of the formulation.