Construction of prediction model for azeotropes by QSPR

Abstract

An azeotrope is a mixture of two or more liquids in such a ratio that its composition cannot be changed by simple distillation. An azeotropic phenomenon is a critical problem in the design of distillation processes. It is important to investigate azeotropes before designing a chemical process. In this study, we proposed a method to construct prediction model for azeotropes through a statistical learning of molecular descriptors and azeotropic data. Using this method,azeotropic properties are calculated very quickly without experimental data. Two statistical models were constructed, 1: predict the presence of an azeotropic point of any binary solution. 2: predict azeotropic composition of binary solution. Compared with the UNIFAC, the constructed models are proved to have comparable predictive accuracy and higher general applicability.