<![CDATA[This study aims to minimize uncertainty in waterflood performance by employing a data-driven workflow that combines the Capacitance Resistance Model (CRM) with Machine Learning. Two CRM variants, CRM-P (Producer-based) and CRM-IP (Injector-Producer-based), are utilized to evaluate interwell connectivity and time constants on three reservoir models: homogeneous, heterogeneous, and a real field scenario (Volve Field). The model is evaluated using R² and Mean Absolute Percentage Error (MAPE) and is compared against the Random Forest and eXtreme Gradient Boosting (XGBoost) techniques. The results indicate that CRM-IP provides more realistic estimates than CRM-P, particularly for response time. XGBoost consistently demonstrates superior prediction accuracy, achieving R² values of 0.76–0.98 and MAPE values of 0.5–10%. Three-dimensional (3D) visualizations of interwell connectivity and streamline analysis strengthen the understanding of fluid flow and sweep efficiency. This further demonstrates that integrating CRM and Machine Learning serves as a decision-support tool for Enhanced Oil Recovery optimization, as evidenced by R² and MAPE analyses that characterize sweep efficiency and the reservoir's capacity to accommodate additional injection.]]>
Copyrights © 2025
