Prediction of the auto-ignition temperature of binary liquid mixtures based on the quantitative structure–property relationship approach

The auto-ignition temperature (AIT) is one of the most important parameters in flammability risk assessment and management in the chemical process. Therefore, in this work, quantitative structure–property relationship approach was employed to estimate the AIT of binary liquid mixtures only based on the information of molecular structures. Various kinds of molecular descriptors were calculated using Dragon 6.0 software after the geometry optimization of molecular structures. Genetic algorithm (GA) was used to select the best subset of descriptors which have a significant contribution to AIT. Two novel models including multiple linear regression (MLR) model and support vector machine (SVM) model were developed based on the GA-selected molecular descriptors. The resulted models showed satisfied goodness-of-fit, robustness and external predictability after the rigorous verification based on appropriate criteria. The MLR model showed great performance with the average absolute error (AAE) of training set and test set being 13.420 °C and 15.076 °C, while the AAE of SVM model was reduced to 5.629 °C and 9.206 °C, respectively. The two optimal models could provide a convenient and effective way to predict the AIT of binary liquid mixtures as well as guidance for the safety design of the chemical process industry.