EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Physics‐Informed MTGODE for Indoor CO 2 Concentration Prediction via Integration of Mass Conservation Law

Yuhan Hu, Suli Zhang, Yanjie Han

Indoor Air · 2026-01 · 2026 (1)

DOI: 10.1155/ina/2063106 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Indoor Air, online erschienen am 23.08.2026. Verfasst von Yuhan Hu und 2 weiteren Autor:innen. Die Arbeit nennt 24 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate prediction of indoor CO 2 concentration is essential for maintaining healthy indoor environments and enabling intelligent ventilation control. Existing data‐driven methods mainly rely on discrete‐time modeling, which limits their ability to characterize the continuous diffusion dynamics of indoor pollutants. Moreover, the absence of physical constraints often leads to cumulative errors during long‐term forecasting, especially under data‐scarce conditions. To address these challenges, this paper proposes a physics‐informed multivariate time‐series graph neural ordinary differential equation (PI‐MTGODE) model. The proposed framework captures spatial dependencies among multiple monitoring nodes through adaptive graph learning and models continuous‐time environmental evolution using Neural ODEs. Furthermore, a derivative‐space physical coupling mechanism is introduced to embed the CO 2 mass conservation law into the Neural ODE vector field, improving the dynamical consistency of long‐term predictions through structural physical constraints. By treating the ventilation decay coefficient k and comprehensive source term f as learnable physical parameters, the proposed model enables unsupervised identification of indoor environmental dynamics. Experimental results show that, compared with the baseline MTGODE, PI‐MTGODE reduces RMSE by 29.7% and 11.7% and decreases the mass conservation residual (MCR) by 39.5% and 66.1% on the evaluated datasets. Under sparse‐data conditions with only 100–400 training samples, the proposed model maintains stable prediction performance and derives physically meaningful parameters consistent with indoor environmental principles. Overall, PI‐MTGODE achieves a balance between prediction accuracy and physical consistency, providing an interpretable modeling framework for indoor air quality prediction and intelligent ventilation control in data‐limited scenarios.

Zitieren

APA

Hu, Y., Zhang, S., & Han, Y. (2026). Physics‐Informed MTGODE for Indoor CO 2 Concentration Prediction via Integration of Mass Conservation Law. Indoor Air, 2026(1). https://doi.org/10.1155/ina/2063106

Vancouver

Hu Y, Zhang S, Han Y. Physics‐Informed MTGODE for Indoor CO 2 Concentration Prediction via Integration of Mass Conservation Law. Indoor Air. 2026;2026(1). doi:10.1155/ina/2063106

BibTeX
@article{Hu2026,
  author = {Hu, Yuhan and Zhang, Suli and Han, Yanjie},
  title = {Physics‐Informed MTGODE for Indoor CO 2 Concentration Prediction via Integration of Mass Conservation Law},
  journal = {Indoor Air},
  year = {2026},
  volume = {2026},
  number = {1},
  doi = {10.1155/ina/2063106}
}
Kostenlos starten →