An MPC-based energy management framework for building HVAC systems considering demand response, thermal comfort, and control stability

The energy consumption of heating, ventilation, and air-conditioning (HVAC) systems poses a significant challenge for buildings, and their operational characteristics can exacerbate grid instability. This study develops a demand-response (DR)-oriented model predictive control (MPC) strategy, termed DPR-PC, which coordinates energy efficiency, indoor thermal comfort, and control stability by optimizing temperature setpoints for the upcoming hour based on a 3-h prediction horizon. An integrated objective function is proposed; four lightweight LightGBM-based prediction models are developed for forecasting indoor temperature Ti, indoor humidity Hi, mean radiant temperature Tr, and power P; and Grey Wolf Optimizer (GWO) is employed for rolling optimization. The controller is implemented within a co-simulation platform that integrates Dymola, EnergyPlus, and Python, which provides a detailed simulation environment for building-HVAC interactions and a feasible pathway for integrating building-HVAC digital twins with AI-based control. Results show that, compared to a constant temperature controller (CTC), DPR-PC achieves 3.77-5.38% energy savings and 3.90-5.45% cost savings during peak periods while maintaining acceptable thermal comfort. Relative to operation at the upper comfort limit of 26 °C, DPR-PC suppresses 52% of the PMV increase and 65% of the PPD increase while retaining 84.9% of the cost-saving potential. Compared with a temperature-deviation-based controller (TDC), DPR-PC reduces median and peak PMV by 19% and 15.7%, respectively. Moreover, the average optimization time of approximately 37 s per control cycle demonstrates the near-real-time feasibility of the proposed controller.