EUVIMEDEUVIMED entdecken
Crossref

Optimal control of subsurface reservoir systems using deep learning methods based on neural operators

Daniil D. Sirota, Kirill A. Gushchin, Egor A. Larionov, Dmitrii Yu. Zverev

Vestnik of Saint Petersburg University. Applied Mathematics. Computer Science. Control Processes · 2025 · 21 (4) · 591-603

DOI: 10.21638/spbu10.2025.410 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Vestnik of Saint Petersburg University. Applied Mathematics. Computer Science. Control Processes, 2025. Verfasst von Daniil D. Sirota und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The paper presents a method for solving problems of optimal control of subsurface reservoir systems based on the use of neural operators. Reservoir systems are complex objects with distributed parameters described by systems of partial differential equations (PDE). Traditional approaches based on the numerical solution of such systems provide high accuracy of reproduction of the dynamics of filtration processes, but are associated with computational complexity. The calculation time for many problems can reach several hours even on highperformance clusters. Thus, the speed of calculations is one of the main factors limiting the use of numerical hydrodynamic models for decision support and optimal control of reservoir systems. Modern deep learning methods, in particular neural operators trained on the results of hydrodynamic modeling, can be considered as an effective and fast alternative to traditional numerical methods in problems of approximating PDE solution operators. Due to their differentiable structure, neural operators support the calculation of functional gradients using an automatic differentiation mechanism. This allows to effectively solve optimization problems, significantly reducing computational costs. The main results of this work include a theoretical estimation of the upper bound of the control error that accounts for the properties of the neural operator, as well as the development of an optimal control algorithm based on the proposed method. The practical significance of the approach is demonstrated through the optimization of control actions for an underground gas storage facility, where a six-order-of-magnitude acceleration in state equation computations was achieved compared to conventional methods.

Zitieren

APA

Sirota, D. D., Gushchin, K. A., Larionov, E. A., & Zverev, D. Y. (2025). Optimal control of subsurface reservoir systems using deep learning methods based on neural operators. Vestnik of Saint Petersburg University. Applied Mathematics. Computer Science. Control Processes, 21(4), 591-603. https://doi.org/10.21638/spbu10.2025.410

Vancouver

Sirota DD, Gushchin KA, Larionov EA, Zverev DY. Optimal control of subsurface reservoir systems using deep learning methods based on neural operators. Vestnik of Saint Petersburg University. Applied Mathematics. Computer Science. Control Processes. 2025;21(4):591-603. doi:10.21638/spbu10.2025.410

BibTeX
@article{Sirota2025,
  author = {Sirota, Daniil D. and Gushchin, Kirill A. and Larionov, Egor A. and Zverev, Dmitrii Yu.},
  title = {Optimal control of subsurface reservoir systems using deep learning methods based on neural operators},
  journal = {Vestnik of Saint Petersburg University. Applied Mathematics. Computer Science. Control Processes},
  year = {2025},
  volume = {21},
  number = {4},
  pages = {591--603},
  doi = {10.21638/spbu10.2025.410}
}
Kostenlos starten →