Prediction of CO2 solubility in deep eutectic solvents using random forest model based on COSMO-RS-derived descriptors

Abstract

This work presents the development of molecular-based mathematical model for the prediction of CO2 solubility in deep eutectic solvents (DESs). First, a comprehensive database containing 1011 CO2 solubility data in various DESs at different temperatures and pressures is established, and the COSMO-RS-derived descriptors of involved hydrogen bond acceptors and hydrogen bond donors of DESs are calculated. Afterwards, the efficiency of the input variables, i.e., temperature, pressure, COSMO-RS-derived descriptors of HBA and HBD as well as their molar ratio, is explored by a qualitative analysis of CO2 solubility in DESs using a simple multiple linear regression model. A machine learning method namely random forest is then employed to develop more accurate nonlinear quantitative structure-property relationship (QSPR) model. Combining the QSPR validation and comparisons with literature-reported models (i.e., COSMO-RS model, traditional thermodynamic models and equations of state methods), the developed QSPR model with COSMO-RS-derived parameters as molecular descriptors is suggested to be able to give reliable predictions of CO2 solubility in DESs and could be used as a useful tool in selecting DESs for CO2 capture processes.

Keywords

• CO2 solubility prediction
• Deep eutectic solvents
• Quantitative structure-property relationship model
• COSMO-RS-derived descriptors
• Random forest