An Intelligent Approach to Predict Minimum Miscibility Pressure of Injected CO2-Oil System in Miscible Gas Flooding

Abstract

Abstract ANN Model was developed utilizing experimentally determined MMP data of 201 reservoir oil and CO2 injected gas. The data bank was randomly divided into training (70%) and testing parts (30%). The conventional statistical measures like coefficient of determination (R2), root mean square error (RMSE) and mean absolute error (MAE) were used to evaluate the predictive efficiency of the model and correlation. Cross-plot of predicted values versus the predicted data was also made to examine the accuracy of developed model. All the important parameters that affect MMP were considered in developing ANN model. These parameters include reservoir temperature, reservoir oil compositions and properties of heptane plus and composition of N2, C1, H2S in the injected CO2 gas stream. The results showed that developed correlation and ANN model can predict the MMP value with high R2, low RMSE and low MAE. The values of R2, RMSE and MAE are 0.9469, 218.7832 and 175.8902 respectively for testing data points. The presented technique can be used to provide an estimate of the MMP in the absence of experimental data and should be utilized in the initial screening of CO2 miscible flooding process. A novel correlation using artificial neural network (ANN) to predict MMP has been developed in this study. The MMP plays an important role in designing the miscible gas flooding processes and to plan appropriate surface injection facilities. MMP is traditionally measured through experimental and non-experimental techniques. The experimental methods are expensive and time consuming and results from currently used correlations vary significantly and hence there is need of reliable, easy and fast prediction technique.