Prediction of auto-ignition temperatures of hydrocarbons by neural network based on atom-type electrotopological-state indices

Abstract

A quantitative structure–property relationship (QSPR) model was constructed to predict the auto-ignition temperature (AIT) of 118 hydrocarbons by means of artificial neural network (ANN). Atom-type electrotopological-state indices were used as molecular structure descriptors which combined together both electronic and topological characteristics of the analyzed molecules. The typical back-propagation (BP) neural network was employed for fitting the possible non-linear relationship existed between the atom-type electrotopological-state indices and AIT. The dataset of 118 hydrocarbons was randomly divided into a training set (60), a validation set (16) and a testing set (42). The optimal condition of the neural network was obtained by adjusting various parameters by trial-and-error. Simulated with the final optimum BP neural network [16-8-1], the results show that most of the predicted AIT values are in good agreement with the experimental data, with the average absolute error being 21.6 °C, and the root mean square error (RMS) being 31.09 for the testing set, which are superior to those obtained by multiple linear regression analysis and traditional group contribution method. The model proposed can be used not only to reveal the quantitative relation between AIT and molecular structures of hydrocarbons, but also to predict the AIT values of hydrocarbons for chemical engineering.