Human thermal risk prediction in indoor hyperthermal environments based on random forest

Abstract

Indoor hyperthermal environments pose a huge threat to the occupants. Predicting workers’ thermal risk in indoor hyperthermal environments contributes to the health and safety at work and the control strategies of indoor hyperthermal environments. Moreover, the quantitative effects of heat acclimation (HA) on the thermal risk are unknown and the weightings of the factors influencing thermal risk need to be determined. To these ends, the simulation experiments were conducted in a climate chamber. The subjects were asked to do treadmill exercise to simulate manual labor. During the experiments, core temperature, heart rate, and subjective perception were recorded. Significant differences were found on the physiological strain index and perceptual strain index before and after HA, which indicate that HA could reduce the thermal risk. Air temperature (Ta), relative humidity (RH), thermal radiation intensity (TRI), clothing insulation (CI), intensity index of physical work (IIPW), and labor hour (LH) were selected as the predictors of the human thermal risk. Given the effect of HA, the non-HA and HA model were built using random forest (RF) with the accuracy of 95.15% and 94.01%, respectively. In addition, the weightings of six contributing factors were determined, and Ta and LH were the most important among them.