Thermal comfort, Personalized comfort, Probabilistic modeling, Bayesian learning
Typical thermal control systems automated based on the use of widely acceptable thermal comfort metrics cannot achieve high levels of occupant satisfaction and productivity since individual occupants prefer different thermal conditions. The objective of this study is to develop environmental control systems that provide personalized indoor environments by learning their occupants and being self-tuned. Towards this goal, this paper presents a new methodology, based on Bayesian formalism, to learn and predict individual occupants thermal preference without developing different models for each occupant. We develop a generalized thermal preference model in which our key assumption, Different people prefer different thermal conditions is explicitly encoded. The concept of clustering people based on a hidden variable which represents each individuals thermal preference characteristic is introduced. Also, we exploited equations in the Predicted Mean Vote (PMV) model as physical knowledge in order to facilitate modeling combined effects of various factors on thermal preference. Parameters in the equations are re-estimated based on the field data. The results show evidence of the existence of multi-clusters in people with respect to thermal preference.