Accurate estimation of permeability reduction resulted from low salinity water flooding in clay-rich sandstones
Abstract
Accurate estimation of permeability reduction in clay-rich sandstones during low-salinity water flooding is critical for optimizing enhanced oil recovery (EOR) strategies and ensuring efficient reservoir management. Traditional methods often rely on costly experiments or simplified empirical correlations, which struggle to capture the complex, non-linear interactions governing this phenomenon. This study introduces a novel data-driven approach utilizing a comprehensive suite of machine learning (ML) methods—including random forest, decision tree, adaptive boosting, ensemble learning, K-nearest neighbors, multilayer perceptron artificial neural networks, convolutional neural networks, and support vector machines—to provide robust predictions of permeability reduction. Methodology of current work, applied to 300 meticulously curated experimental observations, involved rigorous data preprocessing (outlier detection, integrity verification) and k-fold cross-validation to ensure generalizability. The results show that random forest and ensemble learning algorithms delivered the highest predictive accuracy, evidenced by the most substantial coefficient of determination (R2) and minimal error metrics. A sensitivity analysis further clarified that while increasing flooding water salinity and ionic strength leads to a reduction in permeability drop, both the flow rate and the sandstone's clay content exhibit a positive correlation with permeability impairment. This work provides a comprehensive, validated, and highly accurate ML framework specifically tailored for predicting complex permeability alterations, offering a superior alternative to conventional approaches and enhancing decision-making in EOR projects.