Use of hybrid-ANFIS and ensemble methods to calculate minimum miscibility pressure of CO2 - reservoir oil system in miscible flooding process

CO2 injection is believed to be a reliable carbon dioxide sequestration and enhanced hydrocarbon recovery technique. Accurate calculation of minimum miscibility pressure (MMP) is of great importance for the design and implementation of CO2 injection processes. This study involves an innovative machine learning-based algorithm called AdaBoost integrated with the classification and regression tree (AdaBoost-CART) to accurately calculate the CO2-reservoir oil MMP for streams with various compositions. Furthermore, another model is presented on the basis of hybrid-adaptive neuro-fuzzy inference system (ANFIS). The reservoir temperature, composition of injection gas, critical temperature of the drive gas (in average), molecular weight of C5+ fraction in crude oil (MWC5+), and ratio of volatile to intermediate components in crude oil (Vol./Int.) are the independent input parameters considered for the modeling. A large number of data available in technical papers related to intelligent models were selected for comparison purpose. The results reveal the accuracy of the presented Hybrid-ANFIS and AdaBoost-CART models over models selected from the literature. It was also found that the most precise results can be obtained using the developed AdaBoost-CART model as the average absolute relative deviation percent is 0.40. The proposed models can be incorporated in various chemical and petroleum engineering simulators for more accurate and reliable MMP calculations. (C) 2021 Elsevier B.V. All rights reserved.

Automated summary

CO2 injection is a reliable technique for carbon dioxide sequestration and enhanced hydrocarbon recovery, and accurate calculation of minimum miscibility pressure (MMP) is crucial for its implementation. In this study, innovative machine learning algorithms were used to accurately calculate the CO2-reservoir oil MMP, showing promising results compared to existing models. These proposed models can be integrated into various chemical and petroleum engineering simulators for more accurate and reliable MMP calculations.