«Equipment and technologies for oil and gas complex»

Application of the U-FNO neural network in conjunction with recurrent neural networks in hydrodynamic modeling of underground gas storage facilities

UDC: 004.8+004.94:622.691.4
DOI: -

Authors:

¹ Lomonosov Moscow State University, Moscow, Russia
² National University of Oil and Gas "Gubkin University", Moscow, Russia
³ Oil and Gas Research Institute, Russian Academy of Sciences, Moscow, Russia

Keywords: underground gas storage facilities, recurrent neural networks, U-FNO neural network, hydrodynamic modeling, geological modeling, neural network architecture

Annotation:

The use of neural networks in hydrodynamic modeling is a promising approach that can improve the accuracy and speed of predicting physical processes associated with gas storage in underground gas storage facilities. The authors of the scientific work examine the possibility of using the U-FNO neural network in hydrodynamic models of underground gas storage facilities and propose methods for improving the architecture of the neural network. In particular, the possibility of improving the attention algorithm, which is used to select the maximum number of Fourier transform components kmax, is being considered. It is proposed to use an attention layer, which will have trainable matrices and learn weights for individual modes. The space of “time” in hydrodynamic models is considered separately, and the possibilities of using recurrent neural networks, which do an excellent job of processing time series, are explored. To implement a recurrent neural network, the authors of the scientific work solved a synthetic problem similar to the processes of injection and withdrawal in underground gas storage facilities. This synthetic problem will help preparing the neural network architecture for further calculations.

Bibliography:

1. Aziz Kh., Settari E. Matematicheskoe modelirovanie plastovykh sistem / per. s angl. A.V. Koroleva, V.P. Kestnera. – M.: Nedra, 1982. – 407 s.
2. Lie K.-A. An Introduction to Reservoir Simulation Using MATLAB/GNU Octave: User Guide for the MATLAB Reservoir Simulation Toolbox (MRST). – Cambridge University Press, 2019. – 674 p. – DOI: 10.1017/9781108591416
3. Startsev N.I., Mikhaylov N.N. Ispol'zovanie neyronnykh setey v gidrodinamicheskom modelirovanii podzemnykh khranilishch gaza // Oborudovanie i tekhnologii dlya neftegazovogo kompleksa. – 2024. – № 3(141). – S. 49–53.
4. U-FNO – an enhanced Fourier neural operator-based deep-learning model for multiphase flow / Gege Wen, Li Zongyi, K. Azizzadenesheli [et al.]. – 2021. – 37 p. – DOI: 10.48550/arXiv.2109.03697
5. Lu Lu, Pengzhan Jin, Karniadakis G.E. DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators. – 2019. – 22 p. – DOI: 10.48550/arXiv.1910.03193
6. Multipole Graph Neural Operator for Parametric Partial Differential Equations / Li Zongyi, N. Kovachki, K. Azizzadenesheli [et al.] // 34th Conf. on Neural Information Processing Systems (NeurIPS 2020), Vancouver, Canada, Dec. 6–12, 2020. – 17 p. – DOI: 10.48550/arXiv.2006.09535
7. Neural Operator: Graph Kernel Network for Partial Differential Equations / Li Zongyi, N. Kovachki, K. Azizzadenesheli [et al.]. – 2020. – 21 p. – DOI: 10.48550/arXiv. 2003.03485
8. Jiang Zhihao, Tahmasebi P., Mao Zhiqiang. Deep Residual U-net Convolution Neural Networks with Autoregressive Strategy for Fluid Flow Predictions in Large-scale Geosystems // Advances in Water Resources. – 2021. – Vol. 150. – P. 103878. – DOI: 10.1016/j.advwatres.2021.103878