A Reinforcement Learning-Based Ventilation Strategy for Occupant-Centric Thermal Comfort and Energy Efficiency

Indoor environmental quality (IEQ) significantly impacts occupant health, well-being, and productivity, highlighting the importance of effective HVAC management in shared office spaces. However, dynamic occupancy patterns and diverse comfort preferences in shared offices pose challenges, leading to energy waste and dissatisfaction. This study investigates reinforcement learning (RL) for HVAC optimization during transient events in shared offices, focusing on thermal comfort and stress perception. An RL agent learns control policies by interacting with real-time sensor data and surveys conducted in a shared office at Louisiana State University. Participants’ comfort and stress were recorded via surveys and heart rate monitors. The reward function balanced energy efficiency with occupant comfort, enabling dynamic adjustment of airflow and temperature. Preliminary results demonstrate a 70% reduction in energy use compared to rule-based control, while preserving high comfort levels. The integration of RNN-LSTM predictive models with RL decision-making highlights a promising approach for dynamic, multi-occupant HVAC management.