A coupled model for predicting CO2 solubility in aqueous NaCl solution

CO2 solubility is a key parameter to evaluate potential for CO2 storage in saline aquifers. However, existing models for predicting CO2 solubility cannot meet requirements of project designs for CO2 storage in various saline aquifers due to some obvious shortcomings, including complicated calculation processes, low accuracy, poor generalization ability, and narrow application range. Since machine learning can solve complex problems efficiently and there are sufficient data of CO2 solubility in pure water covering wide temperature and pressure (T-P) range, a novel modeling framework is proposed. And then, the above-mentioned framework and machine learning strategy are used to establish an artificial neural network coupled model for predicting CO2 solubility in aqueous NaCl solution. Meanwhile, comprehensive performance of the coupled model is evaluated by considering various evaluation indexes [i.e., mean square error (MSE), coefficient of determination (R2), and mean absolute error (MAE)], natural change trend of CO2 solubility, and leverage approach. The coupled model has better overall performance for predicting CO2 solubility in aqueous NaCl solution (MSE = 0.0041 m2, R2 = 0.991, and MAE < 0.04 m). Additionally, compared with Duan and Menad models, the accuracy and model stability are improved greatly in the coupled model, and its application range is also expanded significantly. Specially, prediction results of the coupled model still have high accuracy in the range of sparse experimental data and it is reliable beyond modeled T-P range. Obviously, the proposed model has strong generalization ability and extrapolation capability. Certainly, this two-step modeling framework can be also applied in the similar problems of uneven data distribution.