Thermal comfort analysis through development of artificial neural network models: An experimental study in Cwa climate

Abstract

This work presents the development of artificial neural network (ANN) models to predict outdoor thermal comfort using the Physiological Equivalent Temperature (PET) index for a college campus in Cwa climate (dry winter humid subtropical climate as per Köppen Climate Classification). Thermal modeling of the study area is exercised using ENVI-met. Twelve different ANN models were developed and tested to evaluate the effectiveness of ANN in predicting the thermal comfort index for the summer and winter months. Two network models were developed using all major meteorological variables. Ten network models were developed using only air temperature, relative humidity, solar radiation, mean radiant temperature, and wind speed respectively. ANN models with all major meteorological variables proved to be effective in estimating close PET values for both summer (R2 = 0.99) and winter (R2 = 0.99). For cases when only one variable is available, network models featuring only air temperature proved effective for both summer (R2 = 0.93) and winter (R2 = 0.92).