Spatiotemporal analysis of thermal comfort profile based on deep learning and refined computational fluid dynamics

Purpose This study aims to provide a detailed spatiotemporal analysis of indoor thermal comfort by introducing a novel approach for reconstructing air temperature and air velocity fields. Design/methodology/approach Strategic deep neural networks and refined computational fluid dynamics (CFD) were employed to reconstruct the air temperature and air velocity fields across five cases with varying window configurations and air diffusers. The predicted mean vote (PMV) and predicted percentage of dissatisfied (PPD) metrics, calculated based on the simulated air temperature and air velocity, were used to assess thermal sensation. Findings The refined CFD simulation approach could reveal a nonuniform distribution of air temperature, while the proposed deep neural networks enable precise and rapid reconstruction of the global temperature and air velocity distribution using only limited measurement data as input and eliminating the need for case-specific model fine-tuning. The spatiotemporal analysis of thermal comfort profiles reveals a nonuniform distribution of thermal comfort within the case office rooms, with cooler areas observed in the central middle of the space. This nonuniform thermal comfort profile may be influenced by design features such as window configurations and air diffusers, with a higher temperature observed near the window facades and the top ceiling. Research limitations/implications The proposed reconstruction methods are scalable to real-world empirical cases with various configurations, enabling accurate and efficient indoor environment modeling, which can further facilitate the design and optimization of real-time heating, ventilation and air conditioning control systems. The simulated spatiotemporal thermal comfort profile can serve as a valuable reference for building managers in climate regulation and space allocation, promoting a comprehensive approach to enhancing energy efficiency while maintaining thermal comfort. Originality/value This research presents a novel approach for reconstructing air temperature fields using deep neural networks, which require minimal data volume and are independent of boundary conditions, coupled with a refined CFD method that reveals the spatiotemporal patterns of air temperature and air velocity rather than relying on statistical mean estimations. Together, these techniques enable more detailed, dynamic and accurate analyses of thermal comfort.