Pioneering oil recovery factor through core‑flooding experiments informed by rock–fluid properties

Résumé fourni par la source

This study presents a novel machine-learning framework for predicting core-scale oil recovery factors by integrating experimentally measured rock–fluid properties with advanced data-driven modeling. A comprehensive dataset of 311 core-flooding experiments including porosity, absolute permeability, interfacial tension, oil–water viscosity ratio, and matrix grain density, was used to capture the nonlinear interactions governing displacement efficiency. Among the evaluated algorithms, the AdaBoost ensemble model achieved the highest predictive accuracy, with strong generalization reflected in its key performance indicators. Model interpretability was enhanced through SHAP analysis, which identified porosity and permeability as the dominant contributors to recovery behavior, consistent with correlation results showing strong positive relationships with RF. The proposed framework provides a transparent and reproducible methodology for linking petrophysical properties to recovery performance and offers practical value for reservoir-engineering workflows by enabling rapid screening and optimization of displacement efficiency.