Modeling the impact of diffusion, Non-Darcy flow, and solubility on core-scale CO2 injection

Reliable forecasting of CO 2 injection behavior demands an accurate description of the fundamental transport and flow mechanisms active within porous media. This work examines how molecular diffusion, non-Darcy flow effects, and CO 2 solubility in the aqueous phase collectively shape oil displacement efficiency and flow characteristics at the core scale. To achieve high-fidelity predictive capability, Gradient Boosting Machines (GBM) were integrated with advanced hyperparameter-optimization strategies. Model robustness was assessed through K-fold cross-validation, while SHAP revealed CO 2 solubility as the most influential factor governing the recovery factor. Among all tested configurations, the GBM-GPO model delivered the strongest performance, achieving an R 2 of 0.9951 and an AARE% of 0.5952, surpassing other machine-learning pipelines. The results demonstrate that hybrid optimization–enhanced ML frameworks not only provide prediction exactness but also yield physically meaningful insights.

Hali tarjima qilinmagan