机器学习驱动碳氢燃料氧化反应自动分类的反应数据集构建方法OA

The ReaxFF MD method is one of the effective tools for exploring the complex reaction mechanisms of hydrocarbon fuels.Although the SRG-Reax method has successfully achieved automatic classification of pyrolysis reactions in ReaxFF MD simulations,it is still unable to handle more complex oxidation processes involving oxygen elements.To expand the application scope of the machine learning-based automatic reaction classification method(SRG-Reax)to hydrocarbon fuel oxidation,this study created an oxidation reaction dataset based on ReaxFF MD simulations of C2 mixture oxidation model(C2H6/C2H4/C2H2/O2).A long-duration oxidation simulation of 2 ns was performed at 3 000 K,and over 120 000 reactions data were obtained.Based on the reaction characteristics between C0 and C1+carbon species,54 oxidation reaction classes were defined under 4 major categories,and 95 955 reactions were manually labeled,which provides a dataset for the training of the automatic oxidation reaction classifier.Additionally,the newly developed oxidation reaction classifier of SRG-reax was used to predict the distribution of reaction types during the oxidation process of n-hexadecane and iso-hexadecane.The automated reaction class predictions of about 100 000 elementary reactions help in dimension reduction of the oxidation reactions and revealing the oxidation pathway differences between the two isomers of n-hexadecane and iso-hexadecane.