Simulation Prediction And Experimental Determination Of CO₂-Oil Minimum Miscibility Pressure

CO₂ Capture, Utilization and Storage technology（CCUS） is the mean way to achieve energy conservation, CO₂ emission reduction and low-carbon green environment. CO₂ enhanced oil recovery（CO₂-EOR） process can realize the utilization and storage of CO₂ underground simultaneously, and becomes the most important and cost-effictive CO₂ utility pattern. The minimum miscibility pressure（MMP） between injected CO₂ gas and crude oil is one of the key parameters during the implementation of CO₂ flooding, and is also the boundary pressure of CO₂ miscible and immiscible floodings. An accurate estimation of CO₂-oil MMP is essential to enabling CO₂-EOR process to be effective and low-cost and yielding significant economical and social benefits.Among the available MMP determination methods, experimental methods are generally difficult and time-consuming to carry out, and high in operation cost. While the theoretical arithmetic techniques for CO₂-oil MMP estimation are usually quick, CO₂ inexpensive and easy to conduct. Therefore, in this work, four different types of theoretical models were proposed to predict or calculate the MMPs in CO₂-EOR processes for both pure and impure CO₂ injection cases. These models include the Winprop fluid phase behavior model, basic artificial neural network（ANN） models, modified ANN models, and the improved numerical correlation. Ten influential parameters that have impacts on MMP were regarded as the input variables of each model, and then obtained the final structure configuration of each developed model or correlation with high prediction accuracy. These parameters are reservoir temperature（TR）, mole fraction of volatile oil components（xvol）, mole fraction of C2-C4 oil components（x C2-C4）, mole fraction of C5-C6 oil components（x C5-C6）, molecular weight of C7+ oil components（MWC7+）, and the gas stream mole fractions of CO₂（y CO₂）, C1（y C1）, H2S（y H2S）, N2（y N2） and hydrocarbons（y HC）. The performances of the developed models were evaluated by comparing its prediction results with（i） experimental results and（ii） predicted results of the available correlations. In addition, the slim tube apparatus was used to measure the MMP value between pure CO₂ and a simulation oil sample. The developed models were applied to predict this MMP value in order to further evaluate their prediction performances using the actual experimental data.Specifically, this dissertation mainly focus on the following five aspects:（1） Selecting a typical crude oil sample and importing the data of its characteristic parameters to the Winprop software to simulate the phase behavior and properties changes of CO₂-oil system in the CO₂ miscible flooding process. Through the fitting of oil PVT parameters, a more accurate fluid phase behavior model can be established. Adjusting the related parameters of the equation of state（EOS） to modify the EOS structure, and then using the multiple contact miscible simulation method to calculate the minimum miscible pressure between CO₂ and crude oil. Besides, the miscible mechanism of CO₂ with the oil sample was also explored.（2） Collecting the MMP datasets and developing two kinds of ANN models, i.e. back-propagation neural network（BPNN） and radial basis function neural network（RBFNN）, to predict the MMP values for both pure and impure CO₂ injection cases simultaneously. Through the proper program to adjust the model parameters to search for the best model construction parameters. The results showed that both BPNN and RBFNN can achieve good prediction performance, all the predicted MMP values have high relevancy with the corresponding experimental ones for both training and testing datasets. Besides, BPNN has a slight better generalization ability and fitness than RBFNN. Furthermore, scalability analysis also showed the excellent prediction accuracy of the established models for any parameter value intervals.（3） Using the space parameter optimization technology of genetic algorithm（GA） and particle swarm optimization algorithm（PSO） to preprocess the initial weights and biases of basic BPNN model, in order to improve the existing local minima or over-fitting problems of BPNN model and further improve the prediction accuracy. Delivery the optimal weights and biases as the initial eigenvalues of BPNN, and built the modified GA-BPNN and PSO-BPNN models. Through the comparison with basic BPNN, two modified models have higher prediction accuracy, and their AAD values reduced by 32.65%（for GA-BPNN） and 22.93%（for PSO-BPNN）, respectively. Comparatively, GA-BPNN has a better generalization ability. By virtue of the sensitive analysis for GA-BPNN model, it was found that the differential changes of MMP alter along with the variations of the influence factors. This is probably should be owed to the fact that the interaction behaviors between the multi-component CO₂ injection gas and crude oil are very complicated and differ from reservoir to reservoir, from condition to condition.（4） Through the analysis about the existing theoretical correlations for MMP prediction, the effects of influence parameters expression form, the correlation type and complexity on prediction accuracy were explored, and then established the improved numerical correlation to reflect the association of MMP and its influencing factors intuitively. The accuracy of the improved correlation was evaluated and validated against experimental data reported in the literature concurrently with those estimated by several well-known correlations. It was found that the improved correlation provided higher prediction accuracy and consistency with literature experimental data than other literature correlations. Even though the improved correlation has a slight higher error than GA-BPNN model, it shared a direct expression about the nonlinear relationship between various influencing factors and MMP, and the correlation structure is simple than GA-BPNN. Besides, the scalability analysis results showed the excellent adaptability of the improved correlations. Overall, the relevance of ten influencing parameters on MMP values followed the order of TR>x C5-C6>MWC7+> xvol>y H2S>y HC>y CO₂>y C1>y N2>x C2-C4.（5） Applying the sand-packed model to measure the MMP value between pure CO₂ gas and a simulation oil sample under the temperature of 40 oC as a practical evidence to further evaluate the prediction performances of the several developed theoretical models. According to the recovery results of a series of displacement experiment under different pressure, the obtained MMP value is 8.2 MPa. Comparison results of theoretical models showed that:（i） the improved correlation has the highest precision（which has a relative deviation（AD） of 4.15% with the measured MMP value）, followed by GA-BPNN model（with AD of 4.88%）, and the last one is RBFNN（with AD of 34.02%）. It should be noted that the simulation sample used in this work has no volatile and C2-C4 light hydrocarbon components in it, but the prediction results of each model, especially the improved correlation and the GA-BPNN model, matched well with the measured MMP value, implying clearly and adequately the high prediction accuracy of the developed theoretical models.