@article{Mao2026, 
author = {Yudong Mao and Peng Zhang and Jun Liu and Jingyue Zhang and Xiwen Feng and Xiaofeng Liang and Zhaosong Fang},
title = {Development of an integrated heat strain index for personalized assessment and forecasting},
year = {2026},
journal = {Building Simulation},
volume = {19},
number = {9},
pages = {2323-2347},
keywords = {occupational heat stress, integrated heat strain index (IHSI), deep learning, personalized monitoring, probabilistic modeling, early warning system},
url = {https://www.sciopen.com/article/10.1007/s12273-026-1492-2},
doi = {10.1007/s12273-026-1492-2},
abstract = {Occupational heat stress threatens worker health and productivity under climate warming, yet common environmental indices cannot fully represent person-specific physiological and perceptual strain. This study developed an integrated heat strain index (IHSI) using high-resolution data from 71 young male participants exposed to nine controlled temperature-speed conditions in an environmental chamber. Environmental, physiological, subjective, and individual variables were integrated through logistic-regression-based scoring, and participant-level validation was used to reduce information leakage from repeated time-series measurements. Results show significant phase-dependent physiological and perceptual changes (p < 0.001). Mean skin temperature continues to rise during recovery, reaching 37.02 ± 0.62 ℃, while subjective discomfort partly decreases, indicating a potential blind spot in self-perception. Partial pressure of end tidal carbon dioxide decreases from 4.45 kPa during preparation to 3.66 kPa during recovery, suggesting that respiratory responses provide additional information beyond thermal and cardiovascular indicators. The IHSI achieves AUC (area under the curve) = 0.8872, average precision = 0.9499, and overall accuracy = 80.6% in participant-level validation. For one-minute-ahead forecasting, CNN-GRU (convolutional neural networks-gated recurrent units) is identified as the best model according to validation-set RMSE, achieving R2 = 0.8808, RMSE = 4.9040, and MAPE = 10.5980% using only heat index, heart rate, age, and body mass index. These findings support the feasibility of a continuous, interpretable, and non-invasive framework for individualized heat-strain monitoring and short-term occupational warning.}
}