Intelligent models to predict the indoor thermal sensation and thermal demand in steady state based on occupants’ skin temperature

Abstract

The correct prediction of thermal sensation is an important factor in energy consumption and satisfaction of occupants. This study examined the effectiveness of six different intelligent approaches for predicting thermal sensation and demand using body temperature data of 615 experiments with an exposure time of 3 h in a controlled office place. At each hour, the temperature of 14 uncovered body points was measured and finally, 1845 temperature data points were extracted. The exposure time had a significant effect on the thermal sensation and insufficient impact on the body temperature. Among all measured temperature data points, four points including middle of forehead (MFH), left cheek (LC), Nose (No), and left hand (LH), were taken as models' inputs. The results indicated that the Gaussian Process Regression (GPR) method offers the best outcomes in prediction of thermal sensation with mean absolute error (MAE) of 0.571 and R2 of 0.84 for the test data points. The MAE and R2 obtained by this model were 0.95 and 0.69, respectively, suggesting that GPR is more accurate and reliable than well-known method PMV. Regarding thermal demand, it was found that the accuracies of the GPR and PMV models were 86% and 69%, respectively. Therefore, the GPR approach is capable of predicting outstanding results for thermal demand compared to the existing models on the basis of environmental factors such as PMV Overall, the present study suggested that intelligent methods based on occupants’ physiological factors estimate the thermal sensation and demand better than available standard methods.