Evaluating the critical performances of a CO2-Enhanced oil recovery process using artificial neural network models

CO2 flooding has attracted particular interest as an effective method in enhancing oil recovery (EOR) because it can significantly improve the oil production and assist in the reduction of carbon emissions from anthropogenic sources by permanently storing the injected gas in geological formations. The applications of hybrid smart tools on evaluating the uncertainties of CO2-EOR projects have been proposed in many studies; however, the model must still be developed in terms of either the architecture or algorithm to more comprehensively and accurately predict the flooding processes. This study aimed to assess the applicability of neural network models (ANN) on forecasting the essential performances of the multi-cycle water-alternating-gas process including the oil recovery factor (RF), oil rate, gas oil ratio (GOR), accumulative CO2 production and net CO2 storage in a specific five-spot pattern scale after a series of injection cycles. A total of 223 numerical samples were simulated and collected to train the networks with the independent variables being the initial water saturation, the vertical-to-horizontal permeability ratio, the WAG ratio, and the duration of each cycle. The results presented excellent accuracy in terms of the oil recovery factor, the cumulative CO2 production and the CO2 storage network model on estimations with overall root mean square errors of less than 3%, while the errors were much greater for unqualified models of the oil rate and GOR. By using the neural networks generated, the optimal injection designs were determined for various reservoir conditions from both technical and economic point of views. Technically, optimized CO2-to-water injection time ratios (WAG) vary according to different reservoir conditions and different targets involving oil recovery and net CO2 sequestration, while the duration of one injection cycle is most favorable at the upper constraint. Nevertheless, with a specific fiscal system, the maximal profit is, however, achieved at the minimal duration and at a WAG less than 1.5 for possible oil prices from 30 $/bbl to 60 $/bbl. Even though the results will differ for dissimilar EOR projects and economic conditions, the proposal of applying traditional ANN models, as in this work, will significantly support the numerical evaluation of CO2-EOR projects compared to conventional procedures.