Prediction of Flash Point of Fatty Alcohol Based on Quantitative Structure-property Relationship(QSPR)

In order to improve the accuracy of predicting flash point of fatty alcohol compounds,a predition method based on the QSPR principle was proposed.Molecular descriptors of 91 kinds of fatty alcohols were calculated with the use of Dragon software and 3 molecular descriptors the most closely relating to the flash points of fatty alcohols were selected from 1 481 descriptors through genetic function approximation(GFA).The models of the 3 molecular descriptors were built by MLR and SVM respectively,and the degree of fitting and predictability were verified using the internal authentication and external validation.Results show that for approach of multivariable linear regression and SVM,the average absolute error(AAE)of the prediction value is 2.870 K and 2.706 K respectively,the root mean square error(RMSE) is 3.451 K and 3.371 K.The SVM model of flash point exceeds the MLR model in accuracy,while the MLR model is more simple and convenient.