Local thermal response differences due to sex and BMI among older adults in warm environments

• Significant differences existed between females and males in terms of torso skin temperature, lower limb skin temperature, and tympanic temperature; • Large individual differences existed in thermal preference and device usage; • Older adults’ thermal sensations were generally consistent with their device usage; • The cross-lagged effect between skin temperature and device usage was the most significant at 5-minute intervals; • When the input variable included the head and wrist skin temperature and device usage, the accuracy of LSTM model was 90%. Individual differences in thermal comfort among older adults may arise owing to sex and body mass index (BMI), leading to different preferred settings for indoor parameters in thermal environments. Meanwhile, providing automatic control strategies for personal comfort systems is very important for the elderly because of their reduced sensitivity and mobility. Thus, this study recruited 26 older adults to join climatic chamber experiments that included the use of local cooling devices at five thermal conditions: 26 °C, 40% relative humidity (RH); 28 °C, 60%; 29 °C, 40%; 32 °C, 50%; and 33 °C, 40%. We introduced three types of local cooling devices: a table fan, an evaporative cooling device, and an air-cooled jacket. During the experiment, we monitored the participants' use of these cooling devices and recorded their skin temperatures and thermal perception votes. Notably, differences in thermal responses between the sexes were more pronounced than those between different BMI groups. There were significant differences in torso and lower limb skin temperatures as well as tympanic temperatures between males and females. There were substantial individual variations in thermal preferences and device usage behaviors. Generally, the participants’ device usage behavior was consistent with their thermal sensations and correlated with the mean skin temperatures recorded 5 minutes prior. We trained the long short-term memory (LSTM) model to predict individual device usage and found it was capable of effectively modeling individual device usage when the input variables included the head and wrist skin temperatures and device speed, achieving an accuracy of 90%.