Application of Gaussian Process Regression to prediction of thermal comfort index

Abstract

In this paper, the theory of Gaussian Process Regression (GPR) was introduced, and the Gaussian Process Regression model was established to predict thermal comfort index. In this model, parameters of activity level, clothing insulation, air temperature, air relative humidity, air velocity and mean radiant temperature were selected as the input vectors, and PMV index was the output vector. The calculated results indicated that the Gaussian Process Regression model had good agreement with those of Fanger's equation. Furthermore, the results of the Gaussian Process Regression model, the BP neural network model and SVM were compared and analyzed, it was concluded that the GP model had relatively higher fitting precision and generalization adaptability. With this model, the requirements of real-time control with PMV index as a controlled parameter in an air-conditioning system could be satisfied.