Prediction of Minimum Miscibility Pressure Between CO₂ and Crude Oil by Integrating Improved Grey Wolf Optimization into SVM Algorithm

CO₂ injection can enhance oil recovery and achieve geological carbon sequestration. The miscibility between CO₂ and crude oil significantly impacts the CO₂ EOR performance. Although the minimum miscible pressure (MMP) can be obtained by slim-tube experiment or slim-tube modeling, it is time-consuming, inconvenient, and complicated. This work aims to enhance the prediction efficiency and accuracy of MMP between CO₂ and crude oil under reservoir conditions by improved and integrated machine-learning approaches. A novel method is proposed to improve the forecasting accuracy and efficiency of the MMP by integrating Grey Wolf optimization (GWO) and improved GWO (IGWO) into the Support Vector Machine (SVM) algorithm. Firstly, data sets are collected and data preprocessing is performed to improve the quality of data sets. Secondly, K-fold cross-validation is applied to enhance the generalization of the model. The MMP is predicted by the SVM algorithm. Thirdly, the MMP prediction can be enhanced by introducing GWO and IGWO algorithms, and the optimal model is investigated to evaluate the effect and convergence of the SVM-GWO and SVM-IGWO algorithms. Fourthly, the predicted MMP and evaluation index (MAE, MAPE) are compared. Finally, the field case study is performed to show the practical potential of the approach. The dominant factors of the MMP include formation temperature (TR), MwC₅⁺ (Molecular weight of pentane plus), MwC₇⁺ (Molecular weight of heptane plus), Volatile (mole fraction of volatile components including N₂ and CH₄), and Intermediate (mole fraction of intermediate components including CO₂, H₂S, and C₂-C₄). The data set is formed by filling 87 groups of missing values using the K-Nearest Neighbor (KNN) algorithm and removing 19 groups of outliers based on the Box-plot detection method. The accuracy is improved by 37.45% and 40.79% using GWO and IGWO based on the MAE compared to SVM. The calculated MAPE shows that the accuracy can be enhanced by 37.79% and 41.29% after adding GWO and IGWO. The SVM-GWO and SVM-IGWO improved the accuracy by 54.16% and 57.12%. The proposed method can accurately determine the MMP between CO₂ and crude oil. The field case study highlights the reliability of the proposed method. The developed method can forecast the MMP between CO₂ and crude oil more efficiently and economically.