Machine learning for the prediction of viscosity of ionic liquid-water mixtures

In this work, a nonlinear model that integrates the group contribution (GC) method with a well-known machine learning algorithm, i.e., artificial neural network (ANN), is proposed to predict the viscosity of ionic liquid (IL)-water mixtures. After a critical assessment of all data points collected from literature, a dataset covering 8,523 viscosity data points of IL-H2O mixtures at different temperature (272.10 K-373.15 K) is selected and then applied to evaluate the proposed ANN-GC model. The results show that this ANN-GC model with 4 or 5 neurons in the hidden layer is capable to provide reliable predictions on the viscosities of IL-H2O mixtures. With 4 neurons in the hidden layer, the ANN-GC model gives a mean absolute error (MAE) of 0.0091 and squared correlation coefficient (R-2) of 0.9962 for the 6,586 training data points, and for the 1,937 test data points they are 0.0095 and 0.9952, respectively. When this nonlinear model has 5 neurons in the hidden layer, it gives a MAE of 0.0098 and R-2 of 0.9958 for the training dataset, and for the test dataset they are 0.0092 and 0.9990, respectively. In addition, comparisons show that the nonlinear ANN-GC model proposed in this work has much better prediction performance on the viscosity of IL-H2O mixtures than that of the linear mixed model. (C) 2022 The Authors. Published by Elsevier B.V.

Automated summary

In this study, a nonlinear model that combines the group contribution method with artificial neural network is proposed to predict the viscosity of IL-water mixtures. The model shows reliable prediction performance and outperforms the linear mixed model.