Application of graph neural networks for interwell-connectivity modeling of injection and production wells

UDK: 4.001.57:622.276.1

DOI: 10.24887/0028-2448-2026-4-94-100

Key words: well connectivity coefficient, graph neural network, attention mechanism, optimization of reservoir pressure maintenance system, dynamic transformation of time scale

Authors: А.А. Gaysin (TatNIPIneft, RF, Almetyevsk; Almetyevsk State Technological University «Petroleum Higher School», RF, Almetyevsk); N.K. Isroilov (TatNIPIneft, RF, Almetyevsk; Almetyevsk State Technological University «Petroleum Higher School», RF, Almetyevsk); R.Kh. Nizaev (TatNIPIneft, RF, Almetyevsk; Almetyevsk State Technological University «Petroleum Higher School», RF, Almetyevsk)

Efficient management of oil field development using reservoir pressure maintenance systems requires accurate assessment of pressure interaction between injection and production wells. This paper presents a method for determination of well connectivity coefficients using Graph Attention Network (GATs). This method implies formalization of well system as directed graph, where nodes represent wells with their performances, while edges represent potential well connections. GAT model is trained to predict production well performance based on historical data (flow rates, pressures, water cut, well spacing), rather than directly calculate the connectivity coefficient. The key advantage of this approach is that target connectivity coefficients are extracted from inherent weights of model attention mechanism, which reflect the contribution of each injection well to pressure change in particular production well. Multi-head architecture enebles simultaneous analysis of fluid-flow effects and spatial factors. The experiments were conducted in two stages. The first stage entailed model training and validation based on simulated data obtained from reservoir simulation model. The second stage included successful test run of this method using actual data from Bobrikovian sediments of Romashkinskoye field. Thus, the developed method offers a tool to obtain physically interpretable estimates of well interactions directly from historical production data. This opens up opportunities for development of intelligent monitoring systems and adaptive optimization of waterflood patterns to ultimately improve the performance of oil field development.

References

1. Zhao-Qin Huang, Zhao-Xu Wang, Hui-Fang Hu et al., Dynamic interwell connectivity analysis of multi-layer waterflooding reservoirs based on an improved graph neural network, Petroleum Science, 2024, V. 21, No. 2, pp. 1062–1080, DOI: https://doi.org/10.1016/j.petsci.2023.11.008

2. Cunliang Chen, Wei Zhang, Baolin Yue, Bin Liu, A new method for quantitative description of dominant channels in high water-cut stage, Improved Oil and Gas Recovery, 2022, V. 7, 7 p., DOI: https://doi.org/10.14800/IOGR.1212

3. Bo Li, Hui Zhao, Botao Liu et al., Graph neural networks and hybrid optimization for water-flooding regulation, Physics of Fluids, 2025, V. 37(8),

DOI: https://doi.org/10.1063/5.0268372

4. Heffer K.J., Fox R.J., McGill C.A., Koutsabeloulis N.C., Novel techniques show links between reservoir flow directionality, Earth stress, fault structure and geomechanical changes in mature waterfloods, SPE-30711-PA, 1997, DOI: https://doi.org/10.2118/30711-PA

5. Gaysin A.A., Nizaev R.Kh., A comprehensive approach to well interference modeling using physically-based graph neural networks (In Russ.), Neftyanaya provintsiya, 2025, No. 4, pp. 251–265, DOI: https://doi.org/10.25689/NP.2025.4.251-265

6. Senin P., Dynamic time warping algorithm review. Information and Computer Science Department University of Hawaii at Manoa Honolulu, USA, 2008, V. 855 (1–23).

7. RD 153-39.0-109-01. Metodicheskie ukazaniya po kompleksirovaniyu i etapnosti vypolneniya geofizicheskikh, gidrodinamicheskikh i geokhimicheskikh issledovaniy neftyanykh i neftegazovykh mestorozhdeniy (Guidelines for the integration and staging of geophysical, hydrodynamic and geochemical studies of oil and oil and gas fields): approved by order of the Ministry of Energy of Russia No. 30 on February 5, 2002, URL: http://techexpert.tatneft.ru/docs/

8. Gaysin A.A., Isroilov N.K. Gilyazov A.Kh., Reservoir pressure calculation in producing wells using machine learning methods (In Russ.), Neftyanaya provintsiya, 2024, No. 3, pp. 123–136, DOI: https://doi.org/10.25689/NP.2024.3.123-136