Prediction of oil-water relative permeability in sandstone and carbonate reservoir rocks using the CSA-LSSVM algorithm

Abstract

Relative permeability of multiphase flow through porous media plays a vital role in the petroleum industry, especially in enhanced oil recovery (EOR) processes. In this study, models were developed based on a combination of Least-Square Support Vector Machine and Couple Simulated Annealing (CSA-LSSVM) algorithm to predict oil-water relative permeability in sandstone and carbonate porous media. Comparing the model to thousands of experimental data resulted in overall squared correlation coefficients (R2) of 0.9866 and 0.9965, and minimum squared errors (MSEs) of 0.0014 and 0.000143 for Kro and Krw, respectively. In addition, Model predictions were found to agree excellently with experimental data. The results of CSA-LSSVM model were compared with some of the well-known mathematical equations including the Purcell, Burdine, Brooks and Corey, Corey, and some empirical correlations for predicting the oil-water relative permeability in a heterogeneous carbonate core sample. The models developed in this study outperform the mathematical equations and empirical correlations yielding an overall squared correlation coefficients of 0.9987 and 0.9994, and minimum squared errors of 0.0003 and 0.0049 for Kro and Krw, respectively. Finally, leverage value was introduced to analyze the whole dataset from which 19 points were diagnosed as possible outlier experimental data.