关键词: iTransformer, 迁移学习, 高温环境, 人体核心温度, 深度学习

Abstract:

To enhance the ability to prevent and control heat stress risks among operators in high-temperature environments, a method for human core temperature prediction integrated with iTransformer with transfer learning was proposed. Large-scale simulated datasets were generated using the joint system thermoregulation model (JOS-3), which served as the foundation for pre-training the iTransformer deep learning model. Subsequently, experimental data from climate chambers were employed to fine-tune the pre-trained model with transfer learning strategies, ultimately realizing the prediction of human core temperature. The results show that the proposed method achieves a mean absolute error (MAE) of 0.103 ℃ and a root mean square error (RMSE) of 0.321 ℃ on the test set. It significantly outperforms Transformer, Reformer, Informer and Long Short-Term Memory (LSTM) methods with the identical transfer learning strategies, as well as deep learning methods trained exclusively on climate chamber experimental data, and the numerical method based on JOS-3.

Key words: iTransformer, transfer learning, high-temperature environments, human core temperature, deep learning