مشخصات پژوهش

Condensate buildup near the wellbore during gas production leads to reduced gas flow and decreased reservoir productivity. This study presents a hybrid approach that integrates experimental wettability alteration using nanoparticles (SiO2, CaCO3, Al2O3) with advanced machine learning (ML) models to enhance condensate recovery in gas condensate reservoirs. Laboratory core flooding experiments were conducted under reservoir-representative conditions, and the resulting datasets were analyzed using six ML algorithms: support vector machine (SVM), random forest (RF), artificial neural network (ANN), linear regression (LR), least squares boosting (LSBoost), and Bayesian methods. Among all models, SVM demonstrated the highest predictive performance across all nanoparticle types. For CaCO3, the model achieved R2 = 0.998, RMSE = 0.473, and MAPE = 0.491% under residual condensate saturation (Re), and R2 = 0.987, RMSE = 0.258, MAPE = 2.458% under residual oil saturation (Sor) conditions. For SiO2, SVM yielded R2 = 0.997, RMSE = 0.569, MAPE = 0.551% (Re) and R2 = 0.991, RMSE = 0.569, MAPE = 0.551% (Sor). For Al2O3, the model obtained R2 = 0.978, RMSE = 1.776, MAPE = 1.682% (Re) and R2 = 0.966, RMSE = 0.725, MAPE = 4.762% (Sor). Overall, CaCO3 nanoparticles provided the highest recovery enhancement, improving condensate recovery by up to 18%. The integration of nanoparticle-assisted EOR with ML-based prediction offers a powerful strategy for optimizing recovery in complex gas condensate reservoirs.