EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Quantifying local and global mass balance errors in physics-informed neural networks

M L Mamud, M K Mudunuru, S Karra, B Ahmmed

Scientific Reports · 05.07.2024 · 14 (1)

DOI: 10.1038/s41598-024-65472-9 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Scientific Reports, online erschienen am 05.07.2024. Verfasst von M L Mamud und 3 weiteren Autor:innen. Beteiligte Einrichtung: Subsurface Science Group, Pacific Northwest National Laboratory, Richland, 99352, WA, USA. lal.mamud@pnnl.gov. Die Arbeit wurde bisher 22-mal zitiert (Crossref) und nennt 34 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: DOE Office of Biological and Environmental Research program, Center for Nonlinear Sciences, Los Alamos National Laboratory, DOE SciDAC4.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Physics-informed neural networks (PINN) have recently become attractive for solving partial differential equations (PDEs) that describe physics laws. By including PDE-based loss functions, physics laws such as mass balance are enforced softly in PINN. This paper investigates how mass balance constraints are satisfied when PINN is used to solve the resulting PDEs. We investigate PINN’s ability to solve the 1D saturated groundwater flow equations (diffusion equations) for homogeneous and heterogeneous media and evaluate the local and global mass balance errors. We compare the obtained PINN’s solution and associated mass balance errors against a two-point finite volume numerical method and the corresponding analytical solution. We also evaluate the accuracy of PINN in solving the 1D saturated groundwater flow equation with and without incorporating hydraulic heads as training data. We demonstrate that PINN’s local and global mass balance errors are significant compared to the finite volume approach. Tuning the PINN’s hyperparameters, such as the number of collocation points, training data, hidden layers, nodes, epochs, and learning rate, did not improve the solution accuracy or the mass balance errors compared to the finite volume solution. Mass balance errors could considerably challenge the utility of PINN in applications where ensuring compliance with physical and mathematical properties is crucial.

Schlagwörter

Stichwörter der Autor:innen

Analytical and numerical solutionsMass balance errorsPartial differential equationsPhysics-informed neural networksPorous media flow

Autor:innen und Institutionen

  • Subsurface Science Group, Pacific Northwest National Laboratory, Richland, 99352, WA, USA. lal.mamud@pnnl.gov.
  • Earth and Environmental Sciences Division, Los Alamos National Laboratory, Los Alamos, NM, 87545, USA. lal.mamud@pnnl.gov.
  • Geology and Geological Engineering, University of Mississippi, University, MS, 38677, USA. lal.mamud@pnnl.gov.
  • Subsurface Science Group, Pacific Northwest National Laboratory, Richland, 99352, WA, USA.
  • Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, Richland, 99352, WA, USA.
  • Earth and Environmental Sciences Division, Los Alamos National Laboratory, Los Alamos, NM, 87545, USA.

Förderung

  • DOE Office of Biological and Environmental Research program
  • Center for Nonlinear Sciences, Los Alamos National Laboratory
  • DOE SciDAC4

Zitieren

APA

Mamud, M. L., Mudunuru, M. K., Karra, S., & Ahmmed, B. (2024). Quantifying local and global mass balance errors in physics-informed neural networks. Scientific Reports, 14(1). https://doi.org/10.1038/s41598-024-65472-9

Vancouver

Mamud ML, Mudunuru MK, Karra S, Ahmmed B. Quantifying local and global mass balance errors in physics-informed neural networks. Scientific Reports. 2024;14(1). doi:10.1038/s41598-024-65472-9

BibTeX
@article{Mamud2024,
  author = {Mamud, M L and Mudunuru, M K and Karra, S and Ahmmed, B},
  title = {Quantifying local and global mass balance errors in physics-informed neural networks},
  journal = {Scientific Reports},
  year = {2024},
  volume = {14},
  number = {1},
  doi = {10.1038/s41598-024-65472-9}
}
Kostenlos starten →