Prediction of critical total drawdown in sand production from gas wells: Machine learning approach

Sand production is a critical issue in petroleum wells. The total drawdown (CTD) an essential indicator of the onset sand production. Although some models are available for CTD prediction, most them proven to lack accuracy or use commercial software. Furthermore, previous correlations have not studied trend analysis verify correct relationships between parameters. Therefore, this study aims build accurate and robust predicting using response surface methodology (RSM) support vector machine (SVM). RSM utilized obtain equation without any SVM model alternative method predict with higher accuracy. This used 23 datasets develop proposed models. strong function vertical depth, cohesive strength, effective overburden stress, transit time correlation coefficients (R) 0.968, 0.963, 0.918, −0.813. Different statistical methods, that is, variance (ANOVA), F-statistics test, fit statistics, diagnostics plots, shown has high more than reported literature. Moreover, ideally follow trend. decreased average absolute percent relative error (AAPRE) by 12.7% compared all published correlations' AAPRE 22.6%–30.4%. lowest 6.1%, highest R 0.995. effects independent variables on displayed three-dimensional plots showed significant interactions.