Optimization of well placement at a field based on a neural operator

UDK: 681.518:622.276.342

DOI: 10.24887/0028-2448-2025-12-107-113

Key words: neural operator, reservoir simulation, machine learning, surrogate modeling, well placement optimization, two-phase filtration

Authors: M.M. Khasanov (Gazprom Neft Companу Group, RF, Saint Petersburg); B.M. Latypov (Ufa State Petroleum Technological University, RF, Ufa); E.V. Yudin (Gazprom Neft Companу Group, RF, Saint Petersburg); R.A. Bondorov (Ufa State Petroleum Technological University, RF, Ufa); N.A. Zyryanov (Saint Petersburg State University, RF, Saint Petersburg);

This paper presents a comprehensive methodology for optimizing the placement of production and injection wells in hydrocarbon reservoirs using surrogate modeling based on neural operators. Traditional well placement optimization requires running a large number of full-scale reservoir simulations, making the process computationally expensive and time-consuming. To address this challenge, the study introduces a fast surrogate model built upon the Latent Neural Operator (LNO), capable of reproducing two-phase flow dynamics under varying geological conditions and various well configurations. The LNO model is trained on a large synthetic dataset which is generated using the industrial simulator tNavigator, comprising 2048 scenarios with randomized well placements. The architecture leverages the Physics-Cross-Attention mechanism to encode physical fields into a compact latent representation and decode them back to high-resolution spatial predictions. This enables accurate approximation of pressure and saturation distributions while remaining invariant to computational grid resolution. The trained model performs a one-year dynamic prediction in approximately 0,052 seconds, enabling rapid evaluation of thousands of scenarios. For optimization, a genetic algorithm is employed to maximize cumulative oil production given a fixed number of wells. The surrogate-based framework successfully identifies configurations that avoid low-permeability zones and improve overall reservoir performance. The results obtained demonstrate that LNO provides predictions comparable to full-physics simulations, although some inaccuracies remain near wellbore area due to error accumulation during autoregressive forecasting. The study highlights the potential of neural-operator-based surrogates to significantly accelerate field development planning and outlines future directions for improving near-well prediction accuracy and integrating surrogate-driven optimization with high-fidelity numerical validation.

References

1. Salasakar S., Prakash S., Thakur G., Recent trends in proxy model development for well placement optimization employing machine learning techniques, Modelling, 2024, V. 5, no. 4, pp. 1808–1823, DOI: https://doi.org/10.3390/modelling5040094

2. Tang H., Durlofsky L.J., Graph network surrogate model for optimizing the placement of horizontal wells for CO2 storage, 2024,

DOI: https://doi.org/10.2139/ssrn.4967577

3. Yudin E.V., Modelirovanie fil’tratsii zhidkosti v neodnorodnykh sredakh dlya analiza i planirovaniya razrabotki neftyanykh mestorozhdeniy (Modeling of fluid filtration in heterogeneous environments for the analysis and planning of oilfield development): candidate of physical and mathematical sciences, Moscow, 2014.

4. Badawi D., Gildin E., Neural operator-based proxy for reservoir simulations considering varying well settings, locations, and permeability fields, 2024,

DOI: https://doi.org/10.48550/arXiv.2407.09728

5. Tang H., Durlofsky L.J., Graph network surrogate model for subsurface flow optimization, Journal of Computational Physics, 2024, V. 512,

DOI: https://doi.org/10.1016/j.jcp.2024.113132

6. Li Z., Kovachki N., Azizzadenesheli K. et al., Multipole graph neural operator for parametric partial differential equations, Advances in Neural Information Processing Systems (NeurIPS), 2020, V. 33, pp. 6755–6766.

7. Wang T., Wang C., Latent neural operator for solving forward and inverse PDE problems, 2024, DOI: https://doi.org/10.48550/arXiv.2406.03923

8. Li Z. et al., Geometry-informed neural operator for large-scale 3D PDEs, Advances in Neural Information Processing Systems (NeurIPS), 2023,

DOI: https://doi.org/10.48550/arXiv.2309.00583

9. Yudin E.V. et al., Approach to a new wells performance planning in a low-permeability reservoirs (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2012, no. 11, pp. 25-29.