Highly Efficient and Accurate Gas-Alkane Binary Mixture Interfacial Tension Equations for a Broad Range of Temperatures, Pressures, and Compositions

Abstract

Summary Gas-alkane interfacial tension (IFT) is an important parameter in the enhanced oil recovery (EOR) process. Thus, it is imperative to obtain an accurate gas-alkane mixture IFT for both chemical and petroleum engineering applications. Various empirical correlations have been developed in the past several decades. Although these models are often easy to implement, their accuracy is inconsistent over a wide range of temperatures, pressures, and compositions. Although statistical mechanics-based models and molecular simulations can accurately predict gas-alkane IFT, they usually come with an extensive computational cost. The Shardt-Elliott (SE) model is a highly accurate IFT model that for subcritical fluids is analytic in terms of temperature T and composition x. In applications, it is desirable to obtain IFT in terms of temperature T and pressure P, which requires time-consuming flash calculations, and for mixtures that contain a gas component greater than its pure species critical point, additional critical composition calculations are required. In this work, the SE model is combined with a machine learning (ML) approach to obtain highly efficient and highly accurate gas-alkane binary mixture IFT equations directly in terms of temperature, pressure, and alkane molar weights. The SE model is used to build an IFT database (more than 36,000 points) for ML training to obtain IFT equations. The ML-based IFT equations are evaluated in comparison with the available experimental data (888 points) and with the SE model, as well as with the less accurate parachor model. Overall, the ML-based IFT equations show excellent agreement with experimental data for gas-alkane binary mixtures over a wide range of T and P, and they outperform the widely used parachor model. The developed highly efficient and highly accurate IFT functions can serve as a basis for modeling gas-alkane binary mixtures for a broad range of T, P, and x.

Machine Learning-Based Interfacial Tension Equations for (H2 + CO2)-Water/Brine Systems over a Wide Range of Temperature and Pressure

“…The widely used parameters are employed to design the equation, ,,, which is as follows γ = c 0 + c 1 T P x + c 2 T 2 x + c 3 T x + c 4 P x + c 5 T P + c 6 x P + c 7 x 2 + c 8 x + c 9 S + c 10 T P S where γ(IFT) is in units of mN/m, T (temperature) is in units of K, P (pressure) is in units of bar, x (H 2 mole fraction) ranges from 0 to 1, and S (salinity) is in units of mol/kg, representing the NaCl concentration. The values of each coefficient are given in Table .…”

“…As in previous studies, , the coefficient of determination ( R 2 ) is used to determine the IFT equation form, which is given as R 2 = ∑ false( X − X̅ false) false( Y − Y̅ false) ∑ ( X − X̅ ) 2 ∑ ( Y − Y̅ ) 2 where X and Y represent the MD-simulated and the ML-predicted IFT values, respectively. We also observe the inverse correlations between R 2 and the root mean square deviation ( RMSD ), indicating that the maximation of R 2 and minimization of RMSD can be achieved simultaneously.…”

“…The workflow of the ML process is shown in Figure . Linear regression is used in the ML process as it is proven to be a viable model for IFT prediction in previous studies. , …”

“…Linear regression is used in the ML process as it is proven to be a viable model for IFT prediction in previous studies. 47,73 The linear regression process is to find the parameter matrix λ fitting the equation as follows…”

“…As in previous studies, 47,73 the coefficient of determination (R 2 ) is used to determine the IFT equation form, which is given as…”