Novel mixture descriptors for the development of quantitative structure-property relationship models for the boiling points of binary azeotropic mixtures

摘要

Binary azeotropes, which contain two chemical constituents, are very common in industry. Understanding azeotropic properties is crucial for effectively separating binary azeotropes. Experimental and theoretical approaches such as ab initio have been used to estimate mixture properties but they are costly and time-consuming. The quantitative structure-property relationship (QSPR) model is a viable alternative approach. The most challenging problem in the QSPR study of mixtures is the computation of numerical descriptors to characterize a mixture. In this study, a series of twenty-two formulas were proposed to derive mixture descriptors from the molecular descriptors of the individual pure compounds. The derived mixture descriptors were employed to establish QSPR models to predict the boiling point of binary azeotropic mixtures. The QSPR model developed was found to be the best with R-train(ing)2 and R-test(2) of 0.92 and 0.90 on the basis of a novel proposed formula, in which some coefficients related to the potential energy contribution were used. Mean absolute errors (MAEs) associated with training and test sets were computed as 9.90 and 10.61, respectively. Twelve out of the twenty-two mixture descriptors resulted in the QSPR models with reasonable statistical qualities and, therefore, were taken into account in the production of an ensemble model via a simple averaging strategy. This caused to improve statistical quality of the final QSPR model. (C) 2019 Published by Elsevier B.V.