Improved XGBoost model for predicting minimum miscibility pressure in CO₂ flooding.

Area of Science:

• Petroleum Engineering
• Chemical Engineering
• Data Science

Background:

• Carbon dioxide-enhanced oil recovery (CO2-EOR) is crucial for increasing oil production and supporting carbon capture, utilization, and storage (CCUS).
• Accurate prediction of Minimum Miscibility Pressure (MMP) is essential for the efficiency of CO2-EOR operations.
• Existing MMP prediction models often lack accuracy or generalizability for both pure and impure CO2 scenarios.

Purpose of the Study:

• To develop a robust MMP prediction model for pure and impure CO2-EOR using an enhanced eXtreme Gradient Boosting (XGBoost) algorithm.
• To investigate the impact of critical temperature of injection gas (Tcm) as a novel variable in MMP prediction.
• To improve the accuracy, interpretability, and generalizability of MMP prediction models.

Main Methods:

• Feature selection using reservoir physics and Pearson correlation analysis.
• Dimensionality reduction via Principal Component Analysis (PCA).
• Hyperparameter optimization of XGBoost using Particle Swarm Optimization (PSO).
• Model interpretability assessment using Shapley Additive Explanations (SHAP).

Main Results:

• The proposed CO2-MMP prediction model achieved high accuracy with R² values of 0.9991 (training) and 0.9845 (testing).
• The model demonstrated superior performance compared to traditional MMP prediction methods.
• SHAP analysis confirmed the model's good explanatory capability and identified key influencing factors.

Conclusions:

• The developed XGBoost-based model provides a transparent, efficient, and generalizable approach for MMP prediction in CO2-EOR.
• This methodology offers valuable insights for optimizing CO2-EOR strategies and supporting cost-effective reservoir development.
• The inclusion of Tcm as a variable enhances the model's predictive power for diverse CO2-EOR applications.