Predicting the equilibrium solubility of CO2 in alcohols, ketones, and glycol ethers: Application of ensemble learning and deep learning approaches

Abstract

In the present work, the solubility of CO2 in solutions comprising various physical solvents (Methanol, Ethanol, Propylene Glycol, n-Pentanol, n-Butanol, n-Propanol, Ethylene Glycol, 2-Butanone, 2-Methoxyethanol, 2-Ethoxyethanol, and Acetone) was modeled using machine learning approaches. In this regard, ensemble learning and deep learning approaches including Bagging Regressor with support vector regression (SVR) as the based estimator (Bagging_SVR), Adaptive Boosting Regressor with Decision Tree as the based estimator (ADA_DT), Decision Tree (DT), and Convolutional Neural Network (CNN) were developed. To develop the models, a broad set of CO2 solubility data (810 experimental data points for 11 alcohol, ketone, and glycol ether solvents) was acquired over various operating temperatures (283.15-373.15 K) and pressures (0.0142-57.38 MPa). In addition, the pressure, temperature, critical properties, and acentric factor of physical solvents were considered as the inputs to the machine learning approaches. To evaluate models’ performance, different graphical and statistical analyzes were conducted. The results revealed that the CNN model is more accurate than other proposed models. Root mean square error (RMSE) and determination coefficient (R2) values for CNN model were obtained as 0.016027 and 0.9943, respectively. Based on the sensitivity analysis, the dissolution of CO2 in physical solvents was highly affected by acentric factors as well as critical temperature and critical pressure. The use of leverage method showed that the used experimental data have high reliability and the CNN model also displayed high credit. Results of this study provide accurate models to estimate the solubility of CO2 in solutions containing physical solvents, which is more straightforward and user-friendly compared to thermodynamic methods with difficult calculations and procedures.