Estimation of minimum miscible pressure in carbon dioxide gas injection using machine learning methods

Abstract In recent decades, Enhanced Oil Recovery (EOR) has emerged as a primary method to increase reservoir oil recovery rates. One of these methods involves injecting miscible and immiscible gases. In miscible gas injection, the minimum miscibility pressure (MMP) is crucial, representing the critical pressure at which these gases can mix effectively with the oil phase. However, accurately determining the minimum pressure required for CO2 to miscible combine with the oil phase has always been a significant challenge. Various methods, including slim-tube tests, analytical models, and empirical correlations, are employed to determine MMP. Nevertheless, experimental measurements are time-consuming and costly. At the same time, mathematical models may yield different estimations. This study introduces an innovative approach using machine learning (ML) techniques to determine CO2-MMP during CO2 flooding. These methods produce reliable models, and advanced CO2-MMP techniques have demonstrated improved performance, significantly reducing time and costs. Furthermore, ML algorithms such as Artificial Neural Networks (ANN), Bayesian networks, Random Forest (RF), Support Vector Machine (SVM), LSBoost, and Linear Regression (LR) were employed to estimate MMP. Input data for these algorithms include CO2, H2S, N2, C1, C2, C3, C4, C5, C6, C7+, MWC5+, MWC7+, T, alongside vol/int. Comparative analysis with experimental MMP data revealed that the Glaso method achieves an accuracy of 0.8749, among the most precise methods, while SVM performed best among the mentioned ML algorithms with an accuracy of 0.986 and RMSE of 0.027.