Personal thermal comfort models based on physiological measurements – A design of experiments based review

Researchers have shown that the physiological-based personal comfort models (PCM) are capable of addressing individual differences as well as transient thermal comfort. Given that physiological-based comfort modeling studies are often very resource-intensive, a well-developed Design of Experiment (DOE) framework could help by optimizing the experimental sequence and use of resources. This study critically reviewed 74 physiological-based PCMs studies and dissected each study based on a DOE framework, dividing the experiments into the experimental procedures, sequences and variables settings. The results indicate that skin temperature, subjects' thermal sensation and air temperature are the leading input variables for PCM. Additionally, the most dominant experiment settings include a 1-min physiological data sampling interval, 10 min interval for reporting thermal vote, a less than 3 h experimental duration, and a fixed clothing level. We found that the subjects' number is independent of the experimental duration (correlation coefficient of 0.0201). Different activity levels and submerging subjects' hands into hot water are also used as thermal stimuli, in addition to the change in air temperature. By applying diverse algorithms, the average predicting accuracy of PCM from selected studies could achieve 85%, and Support Vector Machines (SVM) shows a superior predicting performance. The prominent limitations of the existing studies include insufficient subject numbers, technical restrictions of sensing devices, cumbersome data collection interfaces, improper machine learning algorithms and lack of diversity consideration. Finally, the review suggested that more related studies in this field should be compiled for cross-validation, helping to trade off the most appropriate experiment designs corresponding to the study objectives.