Determination of gas–oil minimum miscibility conditions is one of the important design parameters to improve the displacement efficiency of the hydrocarbon reservoir during enhanced oil recovery with gas injection. In this work, a support vector regression (SVR) model is developed using experimental data to estimate the minimum miscibility pressure (MMP) for various reservoir fluids and injection gases. Experimental MMP data taken from the reliable literature were used as input. Each data point input includes methane and intermediate components mole percent, plus fraction properties and reservoir temperature related to reservoir fluid and CO2, H2S, N2 and intermediate mole fractions, and intermediate properties of the injected gas. Experimental MMP is regarded as the model output. The database contains 135 datasets, from which 125 datasets were used for model development, and the rest were used for model evaluation. Genetic algorithm was implemented to optimize the SVR model parameters. The proposed data-driven model was verified by statistical validation data. The model results illustrate a correlation coefficient (R2) of 0.999. In addition, the SVR results demonstrate the proposed model to be a fast tool and a robust approach to map input space to output features. The SVR model was compared to popular data-driven MMP estimation models as well. This comparison presents an acceptable accuracy relative to this estimation model. Finally, the presented model was evaluated against a comprehensive theoretical model of slim tube compositional simulation on a trusted literature dataset.
Read full abstract