AI-Driven Low Salinity Waterflooding Model for Predicting Oil Recovery in Sandstones

Abstract In recent years, low salinity waterflooding (LSWF) has gained significant attention in China. However, laboratory studies show that LSWF effectiveness in increasing oil recovery is influenced by various factors. Identifying the key factors driving the low salinity effects (LSE) is essential for optimizing the LSWF technology and predicting the related oil recovery. This paper reviewed all experimental data from LSWF research in sandstone reservoirs of China and analyzed the correlations between increased oil recovery and individual/multiple influencing factors. Correlation analysis indicated that LSE is significantly correlated only with wettability change, with no clear correlation with other factors based on single-factor analysis, suggesting that LSE is governed by the coupled effects of multiple parameters. Machine learning (ML) algorithms, including decision trees, random forests, gradient boosting decision trees (GBDT), and feedforward neural networks, were used to predict oil recovery using LSWF. Results showed that GBDT had the best- fitting performance and prediction accuracy with a testing R2 of 0.79. Feature importance analysis using GBDT revealed that permeability and porosity were the most critical factors, followed by water salinity, acid number, and clay content. SHapley Additive exPlanations (SHAP) analysis further revealed positive contributions from permeability, LSW salinity, clay content, acid number, and porosity, while HSW salinity and oil base number had negative effects on LSE. These ML-based findings quantify the key factors controlling LSWF performance and provide guidance for its optimization and experimental validation.