Novel Way to Predict the MMP of a CO2–Oil System Using Stacking Models

Abstract

CO2-enhanced oil recovery (EOR) is an important development method for the third oil recovery stage, which occupies a certain position in carbon capture, utilization, and storage (CCUS). CO2–EOR has two kinds of displacement states in the reservoir, namely, miscible displacement and immiscible displacement, and the recovery of miscibility is far better than that of immiscible flooding. Minimum miscible pressure (MMP) plays a crucial role in whether the CO2–oil system can achieve miscibility, so accurate MMP prediction is required to formulate the reservoir development plan. Traditional methods such as slim tube experiments are expensive and time-consuming. Empirical formulas perform slightly inferiorly in terms of accuracy and range of use. In recent years, machine learning, which uses more, has improved in accuracy, but the performance of this prediction still needs to be further optimized. The work used a stacking approach, one of the ensemble models, to filter and fuse several basic machine learning models to further improve the regression effect of MMP data. First, the correlation analysis and variance inflation factor of the MMP data in the dataset are carried out, and the redundant data are excluded for the correlation and collinearity problems. A total of 147 pretreated MMP data were then regressed using 7 baseline models, whose results were preliminarily screened and combined with empirical formula data to form a new dataset. After that, the final output result is obtained through a stacking model and evaluated. In addition to fitting curves, the results of the Stacking model demonstrate the improvement of the stacking model in MMP prediction from three aspects: mean absolute error (MAE), root-mean-square error (RMSE), and decision coefficient (R2).