Minimum miscibility pressure estimation for CO 2 ‐assisted EOR projects using grey wolf optimized intelligent models

Abstract Accurate prediction of minimum miscibility pressure (MMP) is critical for the success of carbon dioxide enhanced oil recovery (CO 2 ‐EOR) and geological carbon storage projects. While traditional empirical correlations and experimental methods are limited by accuracy, cost, and time constraints, machine learning (ML) offers a promising alternative for robust MMP estimation. This study develops five well‐established ML models, including random forest (RF), support vector regressor (SVR), categorical boosting (Cat), extreme gradient boosting (XGB), and artificial neural network (ANN), and optimizes them using the grey wolf optimizer (GWO) for hyperparameter tuning. These models were trained on 980 samples derived from cell‐to‐cell simulations, incorporating reservoir temperature, fluid composition, molecular weight of C₇ + fractions, and specific gravity as input features. Model robustness validated through 5‐fold cross‐validation, ensuring consistent performance across datasets. All optimized ML models significantly outperformed the conventional correlation, achieving significant prediction accuracy with R 2 values for test data between 0.9918 and 0.9985 compared to Yellig's R 2 of 0.7918. The SVR model demonstrated superior performance with the lowest mean relative error (0.0066) and well‐distributed, symmetric residuals centred around zero. Six independent cases confirmed the superior generalization capability of the SVR model, which achieved the lowest MRE of 3.77%. These results confirm that SVR offers high predictive accuracy with computational efficiency, demonstrating robust, generalizable performance for practical CO 2 ‐EOR MMP prediction.