QSPR Studies for Predicting Flash Points of Alcohols using Group Bond Contribution Method with Back-Propagation Neural Networks

摘要

A quantitative structure-property relationship (QSPR) model was established for the prediction of flash points via artificial neural networks (ANNs) using group bond contribution method. Information of both group property and group connectivity in molecules was contained in the model, and the back-propagation (BP) neural networks which had the high ability of nonlinear prediction were employed. The dataset was composed of 58 alcohol compounds, with experimental flash points values ranging from 11 to 129°C. The molecular structure of each alcohol was characterized by a set of 12 molecular group bonds which were used as input descriptors for model construction. The optimal condition of the neural networks was obtained by adjusting various parameters by trial-and-error. Simulated with the final optimum BP neural networks [16-3-1], the results showed that the predicted flash points for the testing set were in good agreement with the experimental data, with the absolute mean absolute error being 5.62K, and the absolute mean relative error being 1.63％, which were superior to those of traditional QSPR approaches. The model proposed can be used not only to reveal the quantitative relation between flash points and molecular structures of alcohols but to predict the flash points of organic compounds for chemical engineering.