Dynamic personalized thermal comfort Model:Integrating temporal dynamics and environmental variability with individual preferences

Understanding human thermal perception is essential for creating comfortable and energy-efficient indoor environments. In this study, we introduce a dynamic deep learning framework, Thermal Comfort Prediction Model using Long Short-Term Memory (TCPM-LSTM) networks, with Reinforcement Learning (RL) to model and predict personalized thermal comfort under varying environmental conditions. Our proposed Personalized Comfort Model with Reinforcement Learning (PCM-RL) captures temporal dynamics and individual differences in thermal sensation, comfort, and preference. PCM-RL shows about a 13.6 % improvement in average reward when using RL with a pre-trained LSTM (TCPM-LSTM) compared to RL without LSTM. This integrated approach allows the RL agent to make more informed decisions, optimizing comfort based on real-time predictions. Moreover, our framework demonstrates more stable learning behavior, with reduced reward variability across episodes, making it a robust tool for personalized comfort management. This study represents a significant step forward in developing intelligent, adaptive systems that optimize human-centric thermal comfort by providing actionable insights for managing indoor environments effectively.