Modeling the solubility of COS and SO۲ in [Emim][TF۲N] using thermodynamic cubic models and machine learning algorithms

In this paper, the modeling of the solubility of Carbonyl sulfide and sulfur dioxide in [Emim][TF۲N] is studied using the PRSV and PRS equations of state and Panagiotopoulos–Reid mixing rule. The average relative deviation calculated for each equation of state for pure compounds and binary systems represents that both models have comparable abilities in the correlation of experimental VLE data. In addition, three machine learning algorithms including artificial neural network, linear regression, and decision tree regressor were applied to evaluate the performance of statistical algorithms in predicting the mole fraction of the solute in [Emim][TF۲N] binary systems based on temperature and pressure as independent features. According to the calculated coefficient of determination (R۲) of each algorithm, among the three mentioned methods applied for modeling the experimental VLE data, decision tree regressor is the best option to train a model for the solubility of Carbonyl sulfide and sulfur dioxide in [Emim][TF۲N]. Even though artificial neural network algorithm is the most complicated algorithm among these three methods, due to a limited number of data points, it is not able to train a proper model since it overfits data. Modeling processes were carried out by using MATLAB ۲۰۲۱b.

Keywords:

Thermodynamic modeling , Machine learning , Carbonyl sulfide , Sulfur dioxide , Room-temperature ionic liquids