Integrating computational fluid dynamic, artificial intelligence techniques, and pore network modeling to predict relative permeability of gas condensate

Ehsan Zeinedini; Bahram Dabir; Mitra Dadvar

doi:10.1038/s41598-022-24468-z

Integrating computational fluid dynamic, artificial intelligence techniques, and pore network modeling to predict relative permeability of gas condensate

Sci Rep. 2022 Dec 12;12(1):21457. doi: 10.1038/s41598-022-24468-z.

Authors

Ehsan Zeinedini¹, Bahram Dabir², Mitra Dadvar¹

Affiliations

¹ Chemical Engineering Department, School of Material Engineering and Advanced Processes, Amirkabir University of Technology, Tehran, Iran.
² Chemical Engineering Department, School of Material Engineering and Advanced Processes, Amirkabir University of Technology, Tehran, Iran. drbdabir@aut.ac.ir.

Abstract

The formation of gas condensate near the wellbore affects the gas liquid two-phase flow between the pores. It may occur in the path between two pores depending on the thermodynamic conditions of the single-phase gas flow, two-phase gas liquid annular flow or the closed path of condensate in the throat. To model the behavior of gas condensate in a network of pores, relative permeability and naturally pressure drop should be calculated. This study obtained the flow characteristics (pressure drop) between the pores at different physical and geometric conditions using computational fluid dynamics (CFD). CFD is time-consuming, so its results were transferred to an artificial neural network (ANN) model and the ANN model was trained. The CFD was replaced with the ANN model for calculating the pressure drop. In addition, instead of utilizing empirical correlations to compute the accurate value of condensate formed in throats' corners at every time step, the flash calculation using Esmaeilzadeh-Roshanfekr equation of state was performed, and closed throats were specified. This accurately estimates gas and condensate distribution in the pore network. Furthermore, the value of condensate that transferred to the adjacent throats was computed using Poiseuille's law. The results showed that the proposed ANN-based proxy model could promote the calculation speed in gas condensate simulation, considering the dynamic change of relative permeability curves as a function of gas condensate saturation. Also, it was found that the relative permeability obtained by the proposed model is in good agreement with the experimental data. By entering the fractures pattern in the network model and predicting the relative permeability of gas and condensate by the proposed method, the role of fractures in gas condensate production in such reservoirs could be predicted. Dynamic changes due to the relative permeability of gas and condensate as a function of saturation can be entered into the reservoir simulation to optimize inertia and positive coupling phenomena to maximized condensate production in gas condensate reservoir.