<![CDATA[This study evaluates an integrated forecasting framework that combines Capacitance-Resistance Models (CRMP and CRMIP) with ensemble machine learning algorithms (Random Forest and XGBoost) to predict CO₂-Enhanced Oil Recovery performance in the heterogeneous Volve Field. Reservoir simulation was performed using tNavigator with CO₂ injection at 941 tons/day (35 MMSCF/day) over 20 years. The results demonstrate the critical influence of CO₂-specific characteristics, with a determined Minimum Miscibility Pressure of 3299.68 psi and a corresponding oil Swelling Factor of 1.19. Machine learning models, particularly XGBoost, significantly outperformed conventional CRM methods, achieving exceptional accuracy (R² = 0.99-1.00, MAPE = 0.44-2.24%) compared to CRMP/CRMIP (R² = 0.55-0.72, MAPE = 16-23%). The CO₂ injection scenario substantially enhanced oil recovery, achieving a cumulative production of 15.73 MMSTB (RF 20.45%) compared to 9.38 MMSTB (RF 12.19%) for waterflooding, representing a 67.7% improvement and incremental recovery of 6.35 MMSTB. Interwell connectivity analysis revealed dynamic reservoir responses with time constants ranging from 916 to 927 days. The integration of physics-based models with non-linear machine learning algorithms significantly improves prediction accuracy while providing comprehensive insights into reservoir dynamics, allowing for optimal CCUS implementation in heterogeneous reservoir systems.]]>
Copyrights © 2025
