Modeling of CO2 solubility in crude oil during carbon dioxide enhanced oil recovery using gene expression programming

CO2 flooding into the petroleum reservoirs has gained much universal attentions owing to the benefits originated from the reduction of greenhouse gas emission and enhanced oil recovery (EOR). Among the all parameters applied for determining the feasibility of a particular CO2 EOR project especially in the miscible mode, solubility of CO2 has a vital function in the design and simulation of the CO2 injection. CO2 solubility has a prominent contribution to reduction of oil viscosity, IFT reduction, and a rise in swelling of crude oil which leads to the oil mobility increases, and improvement in oil recovery. In the present study, gene expression programming (GEP) as a recently developed and powerful soft-computing technique was utilized for establishing new symbolic CO2 solubility correlations in both dead and live oil systems. Moreover, the prediction capability of the artificial neural network (ANN) was also examined. For this reason, a database of wide-ranging operational conditions was undertaken from the open literature. The parameters involved in both ANN and GEP-based schemes are temperature, saturation pressure, oil molecular weight, and oil specific gravity in dead oil model. In addition to these parameters, the impact of bubble point pressure has been introduced in live oil system. At the next step, these datasets were separated into the two subsets of training group to construct the model and testing group to check the model capability. For assessing the efficiency of the suggested tool, several statistical calculations and graphical illustrations were utilized. Extensive error analysis applied for the newly suggested GEP-based correlations represents satisfactory agreement with highly accurate results of AARD = 0.0378, and R2 = 0.9860 in dead oil, and AARD = 0.0376, and R2 = 0.9844 in live oil. In accordance to the results of this study, the best performance of the extended GEP-based tool in this study was demonstrated compared to the studied literature correlations. The results of trend analysis prove that the proposed model has the best match with the measured datapoints as compared with previously published correlations in both dead and live oil systems. The developed ANN model gives slightly better results than GEP-based correlation in live oil; however, in dead oil the GEP-based strategy is more accurate than the ANN technique. At last, it is found out the GEP-based model can serve as reliable and robust method for rapid and efficient estimation of CO2 solubility in dead and live oil systems.