No abstract

No abstract

No abstract

A neural network-based group contribution method was developed in order to estimate the temperature-dependent surface tension of pure ionic liquids. A metaheuristic algorithm called gravitational search algorithm was employed in substitution of the traditional backpropagation learning algorithm to optimize the update weights of our neural network model. A total of 2307 experimental data points from 229 data sets of 162 different ionic liquid types, such as imidazolium, ammonium, phosphonium, pyridinium, pyrrolidinium, piperidinium, and sulfonium, were collected from the specialized literature. In this database, a wide temperature range from 263 to 533 K, and a wide surface tension range from 0.015 to 0.062 N·m–1, were covered. The input parameters contained the following properties: absolute temperature, the molecular weight of the ionic liquid, and 46 structural groups that composed the molecule. The accuracy of the proposed method was checked using the mean absolute percentage error (MAPE) and the correlation coefficient (R) between the calculated and experimental values. The results show that, for the training phase, our method presents a MAPE = 1.17% and R= 0.998, while for the prediction phase, the method shows a MAPE = 1.29% and R = 0.991. In addition, the relative contribution of each input parameter was calculated from the optimal weights of the network. Also, the effects of the temperature, molecular weight, and cation and anion types on the estimation of the surface tension were analyzed. Finally, the proposed method was compared with other methods available in the literature. All results demonstrated the high accuracy of our method to estimate the temperature-dependent surface tension for several ionic liquid types.

No abstract