Physics-informed neural networks for predictive multiphase flow in REV-scale porous media
Physica Scripta · 26.06.2026 · 101 (25) · 256003
DOI: 10.1088/1402-4896/ae79e6 ↗Kurz gefasst
Fachartikel in der Zeitschrift Physica Scripta, online erschienen am 19.06.2026. Verfasst von Qifeng Li und 2 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
This study develops a physics-informed neural networks model (R-MP-PINNs) for solving multiphase flow in porous media at the representative elementary volume (REV) scale. To address the strong nonlinear coupling between pressure and saturation, the model employs two separate neural networks, each dedicated to one physical field. These neural networks are trained using an alternating optimization strategy with shared variables, which effectively decouples the physical interactions, reduces computational complexity, and improves training stability and efficiency. R-MP-PINNs was applied to investigate several benchmark two-phase flow problems. Results demonstrate that R-MP-PINNs achieves high predictive accuracy and strict physical consistency. Notably, in three-dimensional simulations, the model maintains low-order errors in both pressure and saturation predictions. In summary, R-MP-PINNs provide an efficient, flexible, and reliable solution method for two-phase flow in porous media at the REV scale. This framework not only advances the mechanistic understanding of multiphase transport in porous media but also provides a practical tool for engineering applications.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Li, Q., Zhao, Z., & Lou, Q. (2026). Physics-informed neural networks for predictive multiphase flow in REV-scale porous media. Physica Scripta, 101(25), 256003. https://doi.org/10.1088/1402-4896/ae79e6
Vancouver
Li Q, Zhao Z, Lou Q. Physics-informed neural networks for predictive multiphase flow in REV-scale porous media. Physica Scripta. 2026;101(25):256003. doi:10.1088/1402-4896/ae79e6
BibTeX
@article{Li2026,
author = {Li, Qifeng and Zhao, Ziqing and Lou, Qin},
title = {Physics-informed neural networks for predictive multiphase flow in REV-scale porous media},
journal = {Physica Scripta},
year = {2026},
volume = {101},
number = {25},
pages = {256003},
doi = {10.1088/1402-4896/ae79e6}
}
EUVIMED