Prediction of CO2 solubility in ionic liquids using simplified molecular input line entry system and machine learning methods

As green solvents, ionic liquids (ILs) were highly effective for CO2 absorption. However, lots of possible anion–cation combinations made experimental measurement of CO2 solubility for each IL time- and resource-intensive. Therefore, a machine learning model was used to establish relationships between structures and properties. 19 342 data containing different temperatures and pressures were used in the model, based on the RDKit software package’s simplified molecular input line entry system (SMILES) coded anions and cations. Especially, a new SMILES decoding method with improved character interpretation capabilities was employed. Then, the algorithm’s deep neural network (DNN) and ten sets of cross-validated Random Forest (RF) were used to predict the solubility of CO2. The Pearson correlation coefficient (r), correlation coefficient (R2), root mean square error (RMSE), and mean absolute deviation (MAE) were used as error indicators. The predicted results showed that the model exhibited better performance with the new SMILES decoding methods, with the following error indicators for the DNN and RF: r of 0.986 and 0.983, R2 of 0.972 and 0.968, RMSE of 0.040 and 0.044, and MAE of 0.024 and 0.027, respectively. Therefore, the DNN had higher effectiveness and accuracy than the RF. Moreover, the accuracy of the model was identified using the data from the literature.