Prediction of CO2 Minimum Miscibility Pressure MMP Using MachineLearning Techniques

摘要

Minimum Miscibility Pressure(MMP)is a key design parameter for gas injection projects.It is a physicalparameter that is a measure of local displacement efficiency while subject to some constraints due to itsdefinition.Also,MMP value is used to tune compositional models along with proper fluid description.Ingeneral,CO2 and Hydrocarbon gases are the most common gases used for(or screened for)gas injectionprocesses and due to recent focus to screen for the coupling of CO2-sequestration and CO2-EOR projects.As CO2-Oil phase behavior is quite different than the hydrocarbon gas-oil phase behavior,researchersdeveloped specialized correlations for CO2.Therefore,there is a need for a tool with expanded rangecapabilities for MMP for CO2 gas streams.The only known measurement technique for MMP that iscoherent with its definition is the use of a Slim-Tube which also restricts the amount of data availableeven though there are other alternative techniques presented over the last 3 decades which all suffer fromvarious limitations.Since correlations are inexpensive one of the inexpensive and easy ways to calculatethe MMP,therefore there have been several correlations developed in past based on correlative physics[9],[18],[24],[28],[50],[80],[82] and phase behavior properties of the oil-CO2 mixture [3],[5],[44].This paper present two separate approaches to calculating the MMP of oil during pure CO2 injection,(1)Analytical correlation where the correlation coefficients were tuned using linear SVM [39],[67] and 2)using a hybrid method(combination of random forest regression [11] and proposed correlation)which verynicely captures the dynamic behavior of CO2.The model takes the compositional analysis of oils up toheptane plus fraction,molecular weight of oil,and reservoir temperature as input parameters.Based onstatistical analysis and cross-plots we showed that the performance of the final proposed method is superiorto all the leading correlations [9],[18],[24],[28],[50],[80],[82] and supervised machine learning [55]methods considered in this work [10],[11],[14],[15],[39],[67].The proposed model works for the widestspectrum of MMP from 1000 to 4900 psia which cover the entire range oils within the scope of CO2 EORbased on the screening criteria [54],[75].