Comparison among different modeling approaches for personalized thermal comfort prediction when using personal comfort systems

Abstract

Technologies of personalized thermal comfort prediction (PTCP) and personal comfort system (PCS) are effective for achieving both goals of occupants’ comfort improvement and energy saving in buildings. The modeling of personal comfort is an important basis for PTCP and PCS. This study explored the key issues of the modeling process of personal comfort by conducting experiments in a climate chamber. Twelve male college students participated in the study. Three machine learning algorithms were applied to develop different PTCP models based on infrared-temperature measurement. The results show that the performance of the models can be improved by introducing parameters representing specific information (such as different individuals and PCS usage modes), or by independently modeling the data of different categories. In addition, this study proposes a novel method to optimize model performance by introducing parameters based on manikin test data. Compared with the use of category feature parameters, this method has higher applicability and convenience. Upon comprehensive tasks in this study, the final form of the developed model for thermal sensation prediction achieved an accuracy of 0.88, which indicates the potential of PTCP based on infrared-temperature measurement in diverse scenarios.