This study presents a data-driven framework for energy-efficient and thermal comfort-oriented HVAC control, advancing beyond traditional systems constrained by static occupancy assumptions. A hybrid dual-channel predictive architecture is proposed, integrating real-time occupant dynamics through a synergy of physics-based and data-driven models. The framework employs an energy baseline developed in EnergyPlus and a Predicted Mean Vote (PMV) baseline derived from the Fanger equation, enhanced by behaviour-driven incremental modules that form a dual-increment modelling structure for both energy and comfort prediction. Occupant behaviour is captured using a millimetre-wave (mmWave) radar perception system, enabling adaptive and interpretable modelling of human–environment interactions. The integrated predictive model acts as an objective function evaluator within a Multi-Objective Particle Swarm Optimisation (MOPSO) engine, leveraging pre-trained surrogate models to generate Pareto-optimal control strategies in real time. Experimental validation shows a 76% improvement in prediction accuracy compared to static models, alongside an 8.7% reduction in energy consumption and a 56.2% enhancement in thermal comfort relative to conventional control approaches. The proposed method demonstrates strong potential for sustainable HVAC operation in commercial and institutional buildings, promoting energy conservation and reducing carbon emissions. • A mmWave radar-based method for fine-grained occupant behaviour quantification. • A data-physics dual-channel architecture boosting integrated prediction accuracy. • A surrogate-assisted MOPSO engine achieves significant multi-objective gains.
Li et al. (Sun,) studied this question.