Experimental Determination And Model Study Of CO₂-Oil Minimum Miscibility Pressure

In recent years,oilfield regards CO2 flooding as one of the main measures to enhance oil recovery(CO2-EOR),the technology can achieve at the same time use and sequestration of CO2,which can effectively using the greenhouse gases to achieve "into" for the purpose of energy conservation and emissions reduction and low carbon environmental protection,the minimum miscibility pressure(MMP)is one of the key parameters in the process of CO2 displacement,accurate for CO2 MMP between crude oil and for realization of CO2 miscible flooding,the social and economic benefit is very important.There are many methods to determine MMP,and the most accurate and reliable one is the experimental method.However,the experimental process is time-consuming and expensive,which requires a high precision of the instrument and is susceptible to human factors.The correlation method is simple and clear,but the result is rough and the adaptability is limited.Equation of state method has certain theoretical basis,but it needs to add critical parameters of components,and the calculation is relatively complex and the accuracy is not high enough.Therefore,it is very important to find a method to predict MMP accurately.Firstly,the minimum miscible pressure of pure CO2 injected into 6 groups of crude oil at different temperatures was measured by thin tube displacement experiment.Then,relevant experimental data were collected,and the MMP between injected gas(pure C0O and non-pure CO2)and crude oil was predicted by using the traditional neural network model,the improved neural network model and the minimum miscible pressure calculated according to the PR equation of state.10 parameters affecting MMP were taken as input variables,which were:Reservoir temperature(TR),easy volatile components of crude oil mole fraction(xvol),intermediate component of crude oil,C2-C4 component mole fraction(XC2-C4)and C5+components in crude oil molecular weight(MWC5+),injected gas mole fraction(yCO2 yH2S yNZ yCH4 yHC)as well as the injected gas critical temperature(Tcm),to establish the forecast model of MMP,compare forecast results with the experimental value,verify the accuracy of the model.Finally,six groups of experimental values were predicted by four models,and the prediction performance of each model was evaluated by actual experimental data.Specifically,this paper mainly carries out the following four aspects:(1)The minimum miscible pressure value of pure CO2 gas injected into six groups of crude oil at different temperatures was determined by thin tube displacement experiment,which was used as the practical basis for testing the prediction performance of several different models in this paper.(2)Relevant experimental data of 222 groups(51 groups of pure CO2-crude oil and 171 groups of non-pure CO2-crude oil)were collected,and the radial basis neural network(RBF)and least squares support vector machine(LSSVM)models were used to predict MMP in both cases.The AARD of RBF model(mean absolute relative deviation between predicted value and experimental value)was 8.14%and 7.63%,respectively.The AARD of LSSVM model was 3.18%and 2.83%respectively.lt was found that both the RBF model and the LSSVM model had better training and prediction effect,and the predicted value was highly correlated with the original experimental value.In comparison,the LSSVM model had better prediction effect.(3)The grey Wolf optimization algorithm(GWO)was introduced to combine with RBF and LSSVM to establish GWO-RBF and GWO-LSSVM improved models respectively.The results show that the GWO-RBF model and the GWO-LSSVM model can well predict the minimum miscible pressure between pure CO2 and non-pure CO2 and crude oil.The AARD of GWO-LSSVM model was 1.60%and 2.94%,respectively.By means of Pearson correlation coefficient and Spearman rank correlation coefficient method,the sequencing of 10 miscible pressure sensitive parameters with minimum influence was obtained.Finally,the new model was verified and its effectiveness was evaluated.(4)The experimental data collected in your own forecast MMP correlations and PR state equation to calculate,calculate the result of the two methods and the accuracy of prediction results of the proposed model in this paper,taken together,this paper established intelligent model calculation method of the prediction precision is superior to the traditional relational and PR state equation method,this article 6 groups of experimental data with four kinds of models to predict respectively,thus optimizing the best model,the results showed,the AARD of RBF model was 12.05%,the AARD of LSSVM model is 7.43%,AARD of GWO-RBF model was 6.17%,the AARD of GWO-LSSVM model is 4.47%,among them GWO-LSSVM model prediction effect is best,GWO-RBF is second model,established several kinds of model prediction and experimental measurement with high inosculation,fully shows that the established minimum miscible pressure prediction model has higher prediction accuracy.