Machine learning-based approach to predict thermal comfort in mixed-mode buildings: Incorporating adaptive behaviors

Abstract

Mixed-mode (MM) buildings are designed to provide mechanical air conditioning and natural passive cooling as regulated by occupants. This would enable the potential of shifting the narrow comfort range in HVAC (heating, ventilation and air conditioning) buildings to a wider range similar to NV (naturally ventilated) buildings. Recent studies have provided evidence of higher degrees of thermal adaptation among occupants in MM buildings. However, limited attention has been given to understanding the linkages between these expanded ranges and the specific adaptive behaviors or contextual factors that influence them. This paper aims to investigate the influence of occupants’ adaptive behaviors on thermal comfort in MM buildings. A one-year field study in two MM office buildings with 5096 valid questionnaires was conducted in Chongqing, China, under hot summer and cold winter climatic characteristics by developing machine learning algorithms compared with classic thermal comfort models. Results show that incorporating adaptive behaviors as input variables enhances the performance of machine learning algorithms, leading to improved overall model performance, while the classic thermal comfort index PMV (predictive mean vote) presents the limited accuracy but the best recall in most cases. This paper also demonstrates that some energy-inefficient thermal adaptations were found in MM buildings during the HVAC mode, such as using air conditioning in mild spring and autumn, and frequent window openings during cooling periods of summer. It is therefore valuable for future research to further focus on how MM buildings both incorporate positive features and reduce negative features during the HVAC and NV modes.