Research on Thermal Sensation Prediction in Shoulder Seasons Using Machine Learning Based on Infrared Thermal Imaging

Existing thermal sensation prediction models typically examine the relationship between skin temperature and thermal sensation during cooling or heating seasons. However, due to significant fluctuations in indoor thermal environments during shoulder seasons and considerable individual variation in clothing preferences, traditional thermal sensation prediction models demonstrate poor predictive performance during shoulder seasons. This study aims to investigate the relationship between facial skin temperature and clothing insulation versus thermal sensation under shoulder seasonal conditions and to establish a predictive model for human thermal sensation influenced by clothing insulation. First, facial temperature data under different clothing conditions are collected online using infrared thermal imaging equipment. Subjective thermal sensations are obtained through questionnaires, enabling analysis of the influence of relationships among clothing insulation, facial temperature, and thermal sensation. Subsequently, correlation analysis is used to identify the facial temperature zones closely related to human thermal sensation. Finally, a random forest algorithm is employed to establish a thermal sensation prediction model. Research findings indicate that during shoulder seasons, the left and right cheeks and lips exhibit a higher correlation with thermal sensation. Due to variations in clothing insulation, thermal sensation models based solely on facial temperature characteristics demonstrate lower predictive accuracy and struggle to overcome interference caused by individual clothing differences. After incorporating clothing insulation as a key input feature parameter, the model’s Root Mean Square Error decreased from 0.869 to 0.533, representing a 38.7% improvement in prediction accuracy. This demonstrates that the clothing insulation parameter plays a crucial role in enhancing the precision of human thermal sensation prediction models during shoulder seasons.