Vision-language models for optimising thermal comfort in offices
DOI:
https://doi.org/10.17979/ja-cea.2026.47.13720Keywords:
Thermal comfort, Vision-language models, Building Automation, Artificial intelligence techniques, OptimizationAbstract
To estimate occupants’ thermal comfort sensation in an enclosed space, it is recommended to use the Predicted Mean Vote index. This index is calculated using, among other variables, two variables that depend exclusively on the human component: metabolic rate and the insulation provided by clothing. Generally, these variables are assumed to be static, based on tables included in the standards. In this work, an innovative methodology for optimising the calculation of this index by integrating Vision-
Language Models to estimate these parameters dynamically from real-time images is presented. Specifically, the performance of four recent models of different sizes and costs has been systematically evaluated via the unified OpenRouter API against a
theoretical ground truth obtained through manual validation by a human. The results show that open-source models offer an acceptable performance suitable for local deployments, with zero operational costs and guaranteeing the occupants’ privacy.
References
Borodinecs, A., Palcikovskis, A., Krumins, A., Zajecs, D., Lebedeva, K., 2024. Assessment of hvac performance and savings in office buildings using data-driven method. Clean Technologies 6 (2), 802–813. DOI: 10.3390/cleantechnol6020041
Castilla, M., Álvarez, J. D., Rodr´ıguez, F., Berenguel, M., 2014. Comfort control in buildings. Springer
Choi, H., Jeong, B., Lee, J., Na, H., Kang, K., Kim, T., 2022. Deep-vision based metabolic rate and clothing insulation estimation for occupant centric control. Building and Environment 221, 109345. URL: https://doi.org/10.1016/j.buildenv.2022.109345 DOI: 10.1016/j.buildenv.2022.109345
ISO Standard 7730, 2025. Ergonomics of the thermal environment—Analytical determination and interpretation of thermal comfort using calculation of the PMV and PPD indices and local thermal comfort criteria. International Organization for Standardization, Geneva, Switzerland.
Liu, J., Foged, I. W., Moeslund, T. B., 2022. Clothing insulation rate and metabolic rate estimation for individual thermal comfort assessment in real life. Sensors 22 (2), 619. URL: https://www.mdpi.com/1424-8220/22/2/619. DOI: 10.3390/s22020619
Wei, Z., Calautit, J. K. S., Wei, S., Tien, P. W., 2024. Real-time clothing insulation
level classification based on model transfer learning and computer vision for pmv-based heating system optimization through piecewise linearization. Building and Environment 253, 111277. URL: https://doi.org/10.1016/j.buildenv.2024.111277 DOI: 10.1016/j.buildenv.2024.111277
Zhang, D., Xiong, Z., Zhu, X., 2025. Evaluation of thermal comfort in urban commercial space with vision-language-model-based agent model. Land 14 (4), 786. URL: https://www.mdpi.com/2073-445X/14/4/786 DOI: 10.3390/land14040786
Downloads
Published
Issue
Section
License
Copyright (c) 2026 María del Mar Castilla, Agustín Pérez-Castro, Juana López-Redondo, José Domingo Álvarez
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
