Intrusive and non-intrusive early warning systems for thermal discomfort by analysis of body surface temperature

Buildings consume huge amounts of energy for the thermal comfort maintenance of the occupants. Real-time thermal comfort assessment is both important in the occupants' thermal comfort optimization and energy conservation in the building sector. Existing thermal comfort studies mainly focus on the real-time assessment of the occupant's current thermal comfort. Nonetheless, in the transient thermal environment, the occupant's current thermal comfort is not steady and changes moment by moment. Hence, a prediction error will be elicited if we merely assess the occupant's current thermal comfort. To address this problem, it is crucial to comprehend the occupant's real-time thermal sensation trend in the transient thermal environment. A novel thermal sensation index that directly accounts for an occupant's current thermal sensation trend is investigated in this study. By integrating the novel thermal sensation index into an ordinary thermal comfort model, a novel composite thermal comfort model is derived, which can simultaneously address the occupant's current thermal comfort and current thermal sensation trend. Next, by utilizing machine learning classifications, we propose the intrusive and non-intrusive assessment methods of the composite thermal comfort model by analysis of the skin/clothing temperatures of ten local body parts measured by thermocouple thermometers and upper body thermal images measured by a low-cost portable infrared camera. The intrusive method reached a mean accuracy of 59.7% and 52.0% in Scenarios I and II, respectively; the non-intrusive method reached a mean accuracy of 45.3% and 42.7% in Scenarios I and II, respectively. The composite thermal comfort model provides a thermal discomfort early warning mechanism and contributes to energy conservation in the building sector.