Artificial neural network modelling for solubility of carbon dioxide in various aqueous solutions from pure water to brine

Abstract

Predicting the solubility of CO2 in various aqueous solutions, such as pure water, brackish water, saline water and brine, has become increasingly important because of the interdisciplinary significance of carbon dioxide solubilization, particularly in the energy and environmental fields. Despite experimental difficulties, the solubility of CO2 in various aqueous solutions of single and multi-salt was measured under limited conditions for specific purposes. In this study, artificial neural network (ANN) model was applied to the solubility data of CO2 (2406 experimental data) in salt-dissolved solutions over a wide range of pressures (0.92–712.31 bar), temperatures (273.15–473.65 K), concentrations of electrostricted water molecules (0–90.12 mol/kgw), and overall mole fractions of dissolved salts (0–25.39 mol%), including supercritical condition of CO2. Compared with the conventional thermodynamic models using many parameters, the accuracy of ANN model with only four input variables was comparable for salt solutions and even feasible for brine-rock systems. While the average of the absolute relative deviation of the thermodynamic model with respect to CO2 solubility data in pure water and single-salt solution (1556 data of total 2406 data) was 5.44 %, that of the developed ANN model for the corresponding data was 4.90 %. The developed ANN model presented the capability of dimensional extrapolation in the prediction of CO2 solubility in salt solutions without over-fitting when the thermodynamic-based input variable was used. The model for the solubility prediction of non-polar gases in salt solutions could be a useful approach in complex salt solutions under a wide range of temperature and pressure.