The use of an artificial neural network to estimate natural gas/water interfacial tension

The gas/water interfacial tension (IFT) is an important property that influences many aspects within the petroleum industry, e.g., the vertical distribution of the hydrocarbons and multiphase flow calculations. Laboratory measurement of IFT usually requires an expensive experimental apparatus and a sophisticated interpretation procedure. This paper presents the use of the artificial neural network (ANN) to estimate the IFT in gas/water systems. A total of 956 sets of experimental data consisting of pure methane and synthetic natural gas were acquired from previous literature reports to develop the model. Seven factors were selected as independent variables to estimate IFT using multivariate parametric regression (MPR): temperature, pressure, mole fractions of the gas compositions (CO2, nitrogen, methane, and ethane), and salt (NaCl) concentration in water. A three-layered (7-19-1) ANN trained with the Levenberg-Marquardt back propagation algorithm was used. The mean absolute error, mean percentage error, root mean squared error, and determination coefficient for all of the datasets were calculated to be 0.81 mN/m, 1.97%, 1.25 mN/m and 0.992, respectively, demonstrating the high estimation accuracy and strong generalization capability of the model. The performance of the ANN was further compared with a newly proposed MPR model and three explicit empirical correlations found in previous literature reports. The comparison result suggests that the estimation accuracy can be improved significantly by using ANN compared with these four other correlations.