Enhancing subsurface multiphase flow simulation with Fourier neural operator.

Area of Science:

• Petroleum Engineering
• Computational Science
• Artificial Intelligence

Background:

• Accurate multiphase flow modeling in oil and gas reservoirs is crucial for optimizing hydrocarbon recovery.
• Traditional physics-based simulations face computational challenges due to grid resolution and geological heterogeneity.
• Data-driven surrogate models offer an efficient alternative for solving flow governing partial differential equations (PDEs).

Purpose of the Study:

• To develop and evaluate a Fourier Neural Operator (FNO) for efficient multiphase flow modeling in heterogeneous reservoirs.
• To assess the FNO's capability in predicting reservoir pressure and oil displacement during water injection.
• To improve FNO accuracy and adaptability across different grid configurations.

Main Methods:

• Employing Fourier Neural Operator (FNO) to extract spectral information from reservoir properties for solving coupled porous flow PDEs.
• Investigating two-phase flow dynamics, specifically water injection scenarios.
• Developing a multi-fidelity FNO model to enhance adaptability and accuracy.

Main Results:

• The FNO accurately predicts reservoir pressure distributions for two-phase flow.
• FNO demonstrates sensitivity to local pressure changes near wells, indicating areas for improvement.
• The multi-fidelity FNO shows enhanced adaptability across various grid configurations.
• The FNO achieves a three-orders-of-magnitude speedup compared to traditional PDE solvers.

Conclusions:

• The developed FNO model significantly enhances computational efficiency for reservoir simulations.
• FNOs show potential to replace repetitive physics-based simulations, advancing uncertainty quantification in reservoir performance.
• The multi-fidelity approach improves the robustness and applicability of FNOs in complex reservoir scenarios.