EUVIMEDEUVIMED entdecken
Crossref

Application of neural network modeling to high-order numerical schemes for the convection-diffusion equation

Irina V. Del, Aleksandr V. Starchenko

Vestnik Tomskogo gosudarstvennogo universiteta. Matematika i mekhanika · 2026

DOI: 10.17223/19988621/101/2 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Vestnik Tomskogo gosudarstvennogo universiteta. Matematika i mekhanika, 2026. Verfasst von Irina V. Del und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study presents a monotonic approximation of convective terms for the numerical solution of the convection-diffusion equation using the finite volume method and explicit difference schemes. The proposed approach is based on interpolation with a weighted cubic spline (wSpline). To improve the accuracy of the approximation, an artificial neural network (ANN) and data-based discretization were used (wSplineNN). The optimal ANN architecture and loss function were identified and used in the model development. The performance of the proposed method was evaluated using the problem of impurity transport from an instantaneous point source in both one-dimensional and three-dimensional settings. The results demonstrate the advantages of the data-driven neural-network-based discretization. The finite volume scheme built on data using ANN provided the best MAE and RMSE values on both one-dimensional and various three-dimensional grids compared to the wSpline scheme used without neural network optimization. For the one-dimensional test case, the MAE and RMSE errors were reduced by approximately 40%. For three-dimensional grids of various resolutions, wSplineNN showed lower errors compared to the classical wSpline scheme; in particular, on a 128×128×60 grid, the MAE and RMSE were reduced by approximately 2% and 10%, respectively.

Zitieren

APA

Del, I. V., & Starchenko, A. V. (2026). Application of neural network modeling to high-order numerical schemes for the convection-diffusion equation. Vestnik Tomskogo gosudarstvennogo universiteta. Matematika i mekhanika. https://doi.org/10.17223/19988621/101/2

Vancouver

Del IV, Starchenko AV. Application of neural network modeling to high-order numerical schemes for the convection-diffusion equation. Vestnik Tomskogo gosudarstvennogo universiteta. Matematika i mekhanika. 2026. doi:10.17223/19988621/101/2

BibTeX
@article{Del2026,
  author = {Del, Irina V. and Starchenko, Aleksandr V.},
  title = {Application of neural network modeling to high-order numerical schemes for the convection-diffusion equation},
  journal = {Vestnik Tomskogo gosudarstvennogo universiteta. Matematika i mekhanika},
  year = {2026},
  doi = {10.17223/19988621/101/2}
}
Kostenlos starten →