Abstract: The auto-ignition temperature (AIT) and its intrinsic quantitative relationship with molecular structure of hydrocarbons were studied by Quantitative Structure-Pharmacokinetics Relationship (QSPR). 76 kinds of oxygen-containing derivative of hydrocarbons in the International Electrotechnical Commission (IEC) database were selected as sample sets, 65 kinds were randomly selected as training sets to set up prediction model and 11 kinds as test sets. Genetic algorithm (GA) was used to screen variables, multiple linear regression (MLR) model and support vector machine (SVM) model were established by combining linear and nonlinear methods respectively, and the auto-ignition temperature of 11 oxygen-containing derivative of hydrocarbons were predicted theoretically. Finally, the performance and application fields of the model were evaluated. The results show that eight characteristic molecular descriptors, such as MATS2e, nCOH, Dv, BEHv2, nCHR, GATS1v, IDE and Du, were obtained by GA. The theoretical predicted values of GA-MLR and GA-SVM models were consistent with the experimental values，and the latter model was better. The two prediction models are stable and have strong prediction ability and generalization performance.

Key words: oxygen-containing derivative of hydrocarbons, auto-ignition temperature, quantitative structure-pharmacokinetics relationship, genetic algorithm, multiple linear regression, support vector machine