A practical deep learning model for core temperature prediction of specialized workers in high-temperature environments

Abstract

The health issues of hazardous operations in high-temperature environments are increasing concerns to the public, especially since global warming and extreme weather conditions have made the high-temperature work more frequent and harsher. The abnormal elevation of human core temperature (T_cr) due to high temperatures directly leads to a decline in physiological functions and may trigger various heat-related health issues, which is especially threatening for those working in such conditions. However, continuous real-time T_cr monitoring and prediction are challenging, particularly considering the hazardous operations in extremely hot environments. To address this problem, a non-invasive T_cr prediction model combining a Kalman filter and a long-term sequence forecasting deep learning model was developed. This model leverages monitored skin temperature (T_sk) and heart rate (HR) as input features, enabling personalized real-time T_cr predictions for various groups of specialized operations personnel. The model's accuracy was validated using the data from a series of chamber experiments with 13 participants under ambient temperatures ranging from 34 to 40 °C and T_cr range of 37–39 °C. The optimal prediction results, evaluated by the test set using seven-point T_sk combined with HR, obtain a MAE value of 0.07, a RMSE value of 0.09, and a R² value of 0.93. Additionally, the errors of 95% of all T_cr predictions fell within ±0.17 °C. The proposed model has the advantage of requiring simple input parameters/features and producing high-accuracy predictions, which makes it a practical tool for health monitoring and protection of hazardous operations in high-temperature environments.