A learning machine approach for predicting thermal comfort indices

Human thermal comfort is influenced by psychological as well as physiological factors. Several comfort indices, such as PMV, PPD, TSENS, ET*, DISC, and SET* (see nomenclature) have been developed. These indices attempt to correlate human thermal comfort with environmental conditions. This paper describes the use of a learning algorithm "support vector machine (SVM) learning" for prediction of the thermal comfort indices. The SVM is an artificial intelligent approach that can capture the input/output mapping from the given data. Support vector machines were developed based on the Structural Risk Minimization principle. Different sets of representative experimental environmental factors that affect a homogenous persons thermal balance were used for training the SVM algorithm. The results demonstrate good correlation between SVM predicted values and those obtained from conventional thermal comfort, such as Fanger Model and "2-Node" model. The "trained SVM" with representative data could be easily and more effectively used to predict the indices compared to other conventional estimation methods.