Estimating CO2 solubility in ionic liquids by using machine learning methods

Ionic liquids (ILs) can be used to absorb a greater quantity of carbon dioxide (CO2) by adjusting the ratio of anions and cations. The methods to obtain CO2 solubility in ILs are as follows: Experiment, Equation of State (EOS) and Molecular dynamics simulation (MD). The experiment is time-consuming and labor-intensive, EOS requires critical parameters, and MD calculation is complicated. However, Machine Learning (ML) methods are embraced with high prediction accuracy and convenience by the chemical and petroleum industry. In this work, 1517 solubility data were collected based on 20 different ILs. Three different machine learning methods (Optimization (PSO), Grey Wolf Optimization (GWO) and Sparrow Search Algorithm (SSA) based on Support Vector Machine (SVR)), using acentric factor (ω), critical temperature (Tc), critical pressure (pc) of ILs accompanied by pressure (p), and temperature (T), as input parameters, were employed to predict CO2 solubility in ILs. Meanwhile, three error indexes (namely root mean square error (RMSE), correlation coefficient (R2) and average absolute relative deviation (AARD)), and the duration of time each specified model (PSO-SVR, GWO-SVR and SSA-SVR) takes in order to evaluate the effectiveness of each prediction model. Although the results show that the three different optimization models (PSO-SVR, GWO-SVR and SSA-SVR) have viable prediction effects, the performance of PSO-SVR model has high accuracy and is the lowest time-consuming among them with RMSE and R2 of 0.01881 and 0.9824, respectively. For the SSA-SVR, the estimated RMSE and R2 are respectively 0.01944 and 0.9824 while they are 0.01889 and 0.9822 for the GWO-SVR, respectively. The results imply that all of the models can give reliable predictions on the CO2 solubilities in ILs and the PSO-SVR model performs slightly better than the SSA-SVR and GWO-SVR models.