In this study, a genetic algorithm based back propagation artificial neural network model was developed and used to predict the minimum miscibility pressure, i.e. MMP, for both pure and impure CO2 injection cases. Ten parameters that affecting the MMP were chosen as input variables, while the MMP was selected as output parameter. These parameters were reservoir temperature-TR, mole fraction of volatile oil components-xvol, mole fraction of intermediate oil components-xint, mole fraction of C5–C6 oil components-xC5-C6, molecular weight of C7+ components-MWC7+, mole fraction of CO2 in solvent-xCO2, mole fraction of C1 in solvent-xC1, mole fraction of N2 in solvent-xN2, mole fraction of H2S in solvent-xH2S, and mole fraction of C2–C4 in solvent-xC2–C4. The performance of the newly developed model was evaluated by calculating the deviations between the predicted and the target values, and was compared with four well known correlations in published literature. Through the comparison, it can be found that our new model outperformed those four correlations with the lowest average absolute relative error of 5.51% and mean square error of 7.01%. The influence degrees of each factor on MMP were also analyzed qualitatively and quantitatively by sensitivity analysis. It was found that xint, xC5–C6, and xH2S have positive effects on MMP, while TR, xvol, MWC7+, xCO2, xC1, and xN2 have negative effects on MMP. In addition, the effect of xC2–C4 on MMP can be neglected. Furthermore, the variations of MMP with the increase of each factor were also considered, and it can be found that the slopes of these curves are not constant all the time and change with the variations of the influence factors.
Read full abstract