Prediction of flash points of fatty alcohols based on artificial neural networks

A quantitative structure - property relationship (QSPR) model based on artificial neural networks was established to predict the flash points of fatty alcohols. A set of topological indices was used as molecular structure descriptors to describe the molecular structure characteristics of fatty alcohols. Using the back - propagation artificial neural networks which have the satisfactory nonlinear prediction ability, the correlation between molecular structures and flash points of fatty alcohols was studied with molecular structure descriptors as input parameters and flash point as output one. The results show that the predicted flash points are in good agreement with the experimental data, which are superior to those of conventional group contribution methods. The method proposed can be used to predict not only the quantitative relation between flash points and molecular structures of fatty alcohols but also the flash points of organic compounds for engineering.