A multi-occupants' comfort-driven and energy efficient control strategy of VAV system based on learned thermal comfort profiles

For heating, ventilation, and air-conditioning (HVAC) systems, thermal comfort and energy saving always contradict each other. This article proposes a personalized feedback-data-based learning approach to quantify thermal comfort in a complaint-driven environment control system. We apply a machine learning algorithm named softmax regression to convert user votes into a probability distribution and fit the distribution to data sets for three different comfort conditions (i.e., uncomfortably warm, comfortable, and uncomfortably cold). We have also taken into account the multi-occupants condition under which different users have different thermal preferences and attained an optimized zone-level temperature set-point that is agreeable to all occupants. Co-simulation EnergyPlus/simulink demonstrates the satisfying performance and feasibility of energy saving potentials of the model in real environment control.