Speaker
Description
Accurate evaluations of thermodynamic equilibria are essential for pore-network modeling (PNM) of gas-condensate flow in porous media. However, repeated equation-of-state (EOS) calculations impose a significant computational burden, limiting the feasibility of large-scale, dynamic simulations. This work presents a neural network–based data-driven proxy framework, implemented using JAX, for efficiently approximating thermodynamic phase behavior required in PNM simulations of gas-condensate systems. A custom implementation of the full Peng–Robinson EOS was developed as part of the same framework, serving both as a high-fidelity alternative to the proxy and as the reference model for network training. The proxy network is trained on EOS-based thermodynamic data spanning a representative range of pressures, temperatures, and fluid compositions. To further improve ease of use and efficiency, training is performed on demand and the resulting network parameters are automatically cached for reuse across simulations with compatible thermodynamic conditions. The trained model predicts phase equilibrium with good accuracy while achieving a substantial reduction in computational cost compared to conventional EOS solving. Integration of the proxy network into a dynamic PNM framework enables efficient simulation of multiphase gas-condensate transport, including phase appearance and disappearance at the pore scale. Results demonstrate that the proposed approach preserves the fidelity of predictions while significantly accelerating simulations. The framework provides a scalable and flexible pathway for incorporating complex thermodynamics into pore-scale models, facilitating improved understanding and upscaling of gas-condensate flow in porous media.
The authors thank the technical and financial support of Petrogal Brasil S.A. (Joint Venture Galp | Sinopec) and the promotion of Research, Development and Innovation (R,D&I) in Brazil by the National Agency of Petroleum, Natural Gas and Biofuels (ANP) for the execution of this project.
| Country | Brazil |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
