Toward smart schemes for modeling CO2 solubility in crude oil: Application to carbon dioxide enhanced oil recovery

This paper presents an artificial intelligence-based numerical investigation on the CO2 solubility in live and dead oils for possible CO2-enhanced oil recovery (EOR). A thorough smart modeling was accomplished by utilizing Multilayer Perceptron (MLP) and Radial Basis Function (RBF) neural network predictors integrated with seven vigorous optimization algorithms. Furthermore, Group Method of Data Handling (GMDH) approach was manipulated to achieve explicit mathematical expressions for the scope of the current study. The modeling was performed on a rich source of data derived from the previously published works. Assessments regarding all extended models demonstrated the Absolute Average Relative Error (AARD) ranges of 1.19%-3.47% and 1.63%-3.13% for live and dead oils, respectively. This indicates the prosperousness of all suggested models for anticipating the CO2 solubility in live/dead oil. A comparison between the proposed models indicated the marginally better performance of the MLP-LM (AARD = 1.19%) and MLP-SCG (AARD = 1.63%) in the case of live and dead oils, respectively. Additionally, the implemented models were compared against various published approaches, and the results revealed that the majority of our newly generated models outperform the prior approaches. In addition, the established GMDH-derived correlations were found to be the most truthful in comparison to other explicit literature correlations. These results provide significant insights for understanding the complex physicochemical processes of CO2-EOR and accurately predicting CO2 solubility in live and dead oils in reservoirs.

Automated summary

This paper utilizes artificial intelligence and neural networks for numerical investigation to predict the solubility of CO2 in live and dead oils, showing that the proposed models perform well and outperform previous methods.