Development of ε-insensitive smooth support vector regression for predicting minimum miscibility pressure in CO2 flooding

Abstract

Successful design of a carbon dioxide (CO2) flooding in enhanced oil recovery projects mostly depends on accurate determination of CO2-crude oil minimum miscibility pressure (MMP). Due to the high expensive and time-consuming of experimental determination of MMP, developing a fast and robust method to predict MMP is necessary. In this study, a new method based on ε-insensitive smooth support vector regression (ε-SSVR) is introduced to predict MMP for both pure and impure CO2 gas injection cases. The proposed ε-SSVR is developed using dataset of reservoir temperature, crude oil composition and composition of injected CO2. To serve better understanding of the proposed, feed-forward neural network and radial basis function network applied to denoted dataset. The results show that the suggested ε-SSVR has acceptable reliability and robustness in comparison with two other models. Thus, the proposed method can be considered as an alternative way to monitor the MMP in miscible flooding process.

Keywords

• CO2 flooding
• minimum miscibility pressure
• ε-insensitive smooth support vector regression
• feed-forward neural network
• radial basis function network