A novel method for predicting the flash points of binary mixtures from molecular structures

Abstract

Flash point (FP) is one of the most important parameters used to characterize the potential fire and explosion hazards for flammable liquids. A quantitative structure–property relationship (QSPR) study is conducted to develop mathematical models for predicting the flash points of binary mixtures from only their molecular structures. The modified Simplex Representation of Molecular Structure (m-SiRMS) descriptors were proposed to characterize the molecular features of a total of 288 binary organic mixtures. A set of optimal descriptor subsets that significantly contribute to the FP property were determined by combining the genetic algorithm with multiple linear regression (GA-MLR) and then develop the prediction model. The resulted model is a multiple linear equation with seven SiRMS descriptors. The average absolute error and the root mean square error of the external test set was 3.48 K and 4.73, respectively. The developed model was rigorously validated using multiple strategies, and further extensively compared to the corresponding model developed based on traditional SiRMS descriptors as well as other previously published models. The results demonstrated the robustness, validity and satisfactory predictivity of the proposed model, and the superiority of the proposed m-SiRMS descriptors as well. The proposed method could be reasonably expected to reliably predict the FP of the binary mixtures for engineering.