Applicability of Worldwide CO2 Worldwide Immiscible Flooding and Prediction

Zhang, Na (Missouri University of Science and Technology) | Wei, Mingzhen (Missouri University of Science and Technology) | Bai, Baojun (Missouri University of Science and Technology)

OnePetro 

Abstract

Carbon dioxide (CO2) flooding is a mature technology in oil industry that finds broad attention in oil production during tertiary oil recovery (EOR). After about five decades of developments, there have been many successful reports for CO2 miscible flooding. However, operators recognized after considering the safety and economics that achieving miscible phases is one of big challenge in fields with extremely high minimum miscible pressure (MMP). Compared with CO2 miscible flooding, immiscible flooding of CO2 demonstrates the great potential under varying reservoir/fluid conditions. A comprehensive and high-quality data set for CO2 immiscible flooding is built in this study. Valuable guidelines have been concluded, and production prediction models are established to further assist the applicability of new projects for the first time. Results show that along with the current method in literature to find applicability guidelines, prediction models involved with important operation and production parameters help to increase the accuracy of CO2 immiscible applicabilities. Data involved in this study are checked for independence for feature selection before utilization. We also find that support vector machine could predict the enhanced oil production rate and CO2 injection efficiency better than multiple linear regression method based on the data set. Furthermore, the multiple linear regression method build an excellent model for the prediction of enhanced oil recovery with an accuracy of almost 100%.

Introduction

A prediction model is a tool for decision making and problem solving that has been applied in variety of fields (e.g., medical science [1-3], meteorology [4], transportation [5, 6], business [7, 8], biology [9, 10], and chemistry [11, 12]) for further applicability evaluation. Eagle et al. built a prediction model to accurately estimate the risk of six month mortality after patients have been hospitalized for acute coronary syndrome (ACS), which provides guidance of the intensity of therapy to clinicians in clinical medicine [13]. Gendt et al. established a numerical weather prediction model to help people to make plans for many activities (e.g., farmers to find the best time for harvest; pilots to schedule the safest path, etc. [14]). In a prediction model, prediction accuracy mainly depends on the methodology of prediction and the quality of data that fed into the model, which is one of the crucial indicator to evaluate the effectiveness of models that researchers spare no efforts to pursue as high of an accuracy as possible.