Physics-informed AI reduces operational CO₂ in façade design study

A computational study combining a physics-informed graph attention network with multi-objective optimization identified double-skin façade designs that reduced operational CO₂ emissions while preserving thermal and ventilation functions. Tested in two contrasting climates, the framework reported emissions reductions of 13–23% in cold-dominant Tashkent and 8–20% in hot-arid Najran relative to baseline scenarios. The researchers developed a Physics-Informed Graph Attention Network, or PI-GAT, as a surrogate model within a framework that also included building energy simulations and the multi-objective NSGA-II algorithm.

The approach was designed to explore façade configurations while addressing more than operational carbon alone. PI-GAT was compared with conventional machine-learning methods, including artificial neural networks, random forests and XGBoost. It achieved higher accuracy and greater robustness, particularly under nonlinear and extreme design conditions, which the authors attributed to the inclusion of physical constraints.

These results concern performance within the study’s computational setup; the supplied findings do not establish performance in occupied buildings or through a field trial. The framework’s preferred solutions varied by climate, as shown by the study’s Pareto analysis. In Tashkent, ventilation needed to be regulated more tightly to support carbon-efficient operation.

In Najran, improved ventilation was more important for the selected low-carbon designs. The reported reductions therefore apply to the modeled Tashkent and Najran scenarios and cannot yet be assumed for other climates, building types or operating regimes. Overall, the study presents PI-GAT with NSGA-II as a computationally efficient approach for climate-adaptive façade design, with further validation needed beyond the tested simulations.

Subject of Research: Physics-informed multi-objective optimization of double-skin façade operational carbon performance

Article Title: Physics-informed graph attention networks for multi-objective optimization of double skin façades: a climate-responsive framework for operational carbon reduction

Article References: Hussien, S. A., Samandarov, B., Bakhritdinov, F., Matkarimov, S., Abdullayeva, I., Ibragimova, S., & Benti, N. E. (2026). Physics-informed graph attention networks for multi-objective optimization of double skin façades: a climate-responsive framework for operational carbon reduction. Scientific Reports. https://doi.org/10.1038/s41598-026-75142-7

Image Credits: AI Generated

DOI: 10.1038/s41598-026-75142-7

Keywords: double-skin façades, operational CO₂ emissions, physics-informed AI, graph attention networks, multi-objective optimization, building energy simulation, ventilation, heat transfer, climate-responsive design, machine learning, Physics-informed, graph