Pure Co2-Oil system minimum miscibility pressure prediction using optimized artificial neural network by differential evolution

Abstract

Miscible CO2 flooding is one of the most attractive enhanced oil recovery options thanks to its microscopic efficiency improvement. A successful implementation of this method depends mainly on the accurate estimation of minimum miscibility pressure (MMP) of the CO2-oil system. As the determination of MMP through experimental tests (slim tube, and rising bubble apparatus (RBA)) is very expensive and time consuming, many correlations have been developed. However, all these correlations are based on limited set of experimental data and specified range of conditions, thus making their accuracies questionable. In this research, we propose to build robust, fast and cheap approach to predict MMP for pure CO2-oil by applying hybridization of artificial neural networks with differential evolution (DE). DE is used to find best initial weights and biases of neural network. Four parameters that affecting the MMP are chosen as input variables: reservoir temperature, mole fraction of volatile-oil components, mole fraction of intermediate-oil components and molecular weight of components C5+. 105 MMP data covering wide range of conditions are considered from the published literature to establish the model. The obtained results demonstrate that our approach outperforms all the published correlations in term of accuracy and reliability