Abstract
CO2 flooding is an efficient technique for enhancing oil recovery and carbon sequestration, where accurate estimation of the minimum miscibility pressure (MMP) is crucial. Conventional experiment and numerical simulations are time-cost and often inaccurate. Therefore, it is essential to develop an efficient and accurate model for MMP prediction. In this work, we introduced a genetic algorithm-optimized random forest (GA-RF) model using 260 samples with eight input features, including reservoir temperature (TR), molecular weight of C5+ pseudo-components (MWC5+), ratio of the mole percent of volatile to intermediate fractions (vol./int.), and mole fractions of injected gas (CO2, H2S, CH4, C2–C4, and N2). Four algorithms-random forest (RF), multilayer perceptron (MLP), support vector regression (SVR), and gradient boosting decision tree (GBDT) are compared. RF performs best and is further optimized by GA. The resulting GA-RF model outperforms traditional methods, with prediction accuracy of 0.99 and a root mean square error of 0.74. The model was applied to CO2-hydrocarbons system from different low-permeability reservoirs in China. Results show that the GA-RF model demonstrated an average relative error of 3.43%. These findings highlight that the GA-RF model has great potentials for predicting MMP during CO2 flooding and improving CO2 miscible flooding efficiency.