Prediction of flash point of hydrocarbon by electrotopological state indices

A quantitative structure-property relationship (QSPR) model using electrotopological state indices was created to predict the flash points of hydrocarbons. These indices were easily to calculate and also had a good discrimination ability for isomers. Electrotopological state indices for atom type (ETSI) of 116 hydrocarbons were calculated and used to describe the structure characteristics. The quantitative relation between flash points and molecular structures of hydrocarbons was studied by the linear regression analysis method and artificial neural network approach, respectively. For the training set, the average relative deviations between the experimental and predicted values of flash point were 3.8% by the linear regression analysis method and 2.7% by the neural network approach, while for the testing set, they were 3.1% and 2.0%, respectively. The results showed that the predicted flash points were in good agreement with the experimental data whether by the linear regression analysis method or by the neural network approach, which were superior to those of traditional group contribution methods. The method proposed can be used to predict the flash points of organic compounds based on molecular structures for chemical engineering.