Localized heating can improve thermal comfort in cold environments while reducing energy use relative to whole-space heating; however, robust control remains challenging under rapidly varying outdoor conditions. This study develops an adaptive thermal-comfort prediction model for infrared localized heating to enable personalized spatiotemporal control. A total of 3243 observations were collected from 64 participants using a catalytic-combustion radiant heater, combining subjective thermal votes with multi-site skin-temperature measurements across a range of operating conditions. The results demonstrate that localized heating effectively alleviates cold discomfort, increasing mean skin temperature (MST) by approximately 0.4–1.0 °C. Continuous heating is recommended when the ambient temperature ranges from 0 to 11 °C, as it achieves a thermal satisfaction rate of approximately 65%. Above 11 °C, intermittent operation maintained MST within 30–33 °C and achieved higher satisfaction (around 72%), while reducing the risk of local overheating and improving energy efficiency. To reduce dependence on detailed heat-transfer modeling, five machine learning classifiers were trained and compared; the Random Forest model performed best. After hyperparameter optimization, the model achieved an accuracy of 0.84, F1 scores of 0.83–0.85 across three classes, and AUC values above 0.93, indicating strong discrimination and generalization. These findings provide a foundation for efficient and personalized intelligent control of localized heating in cold outdoor settings.
Jiang et al. (Wed,) studied this question.