Can occupant-centric control be generalizable? A practical occupancy-based framework for heating and cooling setpoint optimization in commercial buildings

This study proposes a practical framework that captures the stochastic nature of occupancy and integrates it into HVAC control sequences. It combines Kernel Density Estimation (KDE), Random Forest (RF), and Cumulative Relative Frequency (CRF) to predict arrival and departure times and occupancy durations. The framework enhances HVAC scheduling and setpoint selection through four operation modes: occupied, unoccupied, standby, and room standby. Performance is evaluated through a two-step approach, including occupancy modelling assessment and simulation-based analysis across multiple datasets and climate zones, benchmarked against a traditional schedule-based system using an EnergyPlus small office prototype. Results demonstrate strong generalizability, achieving energy savings of up to 14% while indicating minimal comfort impact. Heating-dominated climates show higher energy-saving potential, while slightly higher deviations occur in hot climates but remain acceptable. Overall, energy savings are primarily driven by occupancy patterns, enabling delayed start-ups, earlier shutdowns, and adaptive setpoint control.