A new approach to characterize CO2 flooding utilizing artificial intelligence techniques

The use of carbon dioxide in miscible flooding has been considered as one of the most effective techniques for enhancing oil production. The flooding efficiency is an extreme function of the minimum miscibility pressure (MMP), therefore, searching for a quick and rigorous method to determine MMP is highly needed. Slim tube experiments are normally used to measure the minimum miscibility pressure. However, such experiments are time-consuming and very costly. Different correlations have been developed to determine the MMP during CO2 injection process. These empirical equations are not widely applicable and might produce severe estimation errors, because they are developed based on limited experimental results. This paper proposes a new technique to evaluate the CO2 flooding and minimize the uncertainties of using numerical approaches. The objective of this work is developing a reliable model to predict the MMP during CO2 flooding. Actual case studies for flooding heterogeneous and anisotropic reservoir were utilized to generate the MMP model, more than 140 data points were used to construct and evaluate the proposed model. Several artificial intelligence techniques were studied to estimate the CO2-MMP for a wider range of conditions. The developed models investigate the effect of API gravity, fluid composition, and injected gas composition on the performance of CO2 flooding operation. The CO2-MMP was estimated using different artificial intelligence techniques including; radial basis function network, artificial neural network, generalized neural network and adaptive neuro-fuzzy inference system. The wellbore condition and reservoir parameters were used to provide an accurate and quick prediction for the flooding performance. Sensitivity study was conducted to optimize the model parameters. Then, the optimized artificial neural network model was utilized to extract an empirical equation. The developed equation was verified using actual field data an acceptable average absolute percentage error (AAPE) of 6.6% was obtained. In addition, the developed CO2-MMP model was compared with different determination approaches. It is found that, the proposed technique outperforms the current CO2-MMP models. This work would afford an effective approach to characterize the CO2-flooding for complex reservoirs, also improve the prediction performance of commercial software, which leads to a better production management in the particular CO2-operations.