A self-contained model-based predictive control framework for HVAC systems using sequential proactive system identification

• Proactive system identification enables automated inverse model training. • Disturbances representing solar and casual heat gains are estimated as time-variant parameters. • SARIMA models forecast disturbances for MPC. • MPC shifts HVAC loads away from on-peak electricity periods. • Heating and cooling costs reduced by 12% and 15%, respectively. This paper presents a self-contained model predictive control (MPC) framework for small commercial HVAC systems that integrates proactive system identification, short-term disturbance forecasting, and receding-horizon optimization. A proactive system identification sequence is designed to create operating conditions suitable for inverse model training and to enable systematic estimation of zone thermal parameters. Unmeasured disturbances arising from solar and casual heat gains and ventilation loads are forecast using seasonal autoregressive integrated moving average (SARIMA) models, which are incorporated directly into the MPC formulation. The proposed approach is demonstrated using EnergyPlus as an emulation environment on a three-zone small commercial building equipped with rooftop units. Results show that the MPC maintains indoor temperatures within comfort bounds during occupied hours while strategically exploiting thermal mass through preheating and precooling to shift loads away from on-peak electricity periods. This yields annual electricity cost reductions of 12% for heating and 15% for cooling, alongside reductions in total heating and cooling energy use of 3% and 12%, respectively. Practical considerations related to model retraining frequency, interzonal heat transfer, and system configuration are discussed, and pathways for extending the approach to larger multi-zone systems and field deployment are identified.