Predicting Rate Constants Of Alkane Reactions Using Machine Learning

Alkanes are important components of fossil fuels,understanding their chemical reaction kinetics is essential for the development of reliable chemical kinetic models and engine design.The reaction rate constant is one of the most important chemical reactions and plays an important role in simulating the combustion reaction network.If the structure of reactants is similar,the chemical reactions of the same type usually share the same mechanism.Based on this premise,we try to use a neural network model to predict rate constants of alkane reactions.In this study,we use machine learning methods to construct temperature-dependent reaction rate constant models for three types of alkane element reactions,including C_nH_2n+2+H→ C_nH₍2n+10+H₂,C_nH_2n+2+O→C_nH_2n+1+OH and C_nH_2n+2+OH→C_nH_2n+1+H₂O reactions,the predicted values of the rate constants for these three types of reactions-are obtained.The results show that the neural network model constructed in this work has good stability in predicting rate constants,and most of the results are better than the transition state theory.This article contains 6 chapters,the main contents are as follows:Chapter 1 introduces foreign and domestic research backgrounds and current research status of the project in this article.Chapter 2 introduces the theoretical methods used in this article,including the calculation of descriptors and the construction of neural networks.Chapter 3 studies the use of machine learning to predict the rate constant of the reaction of alkanes with hydrogen atoms.The neural network model was trained using the reaction rate constants of 8 reactions of alkanes with hydrogen atoms.The results show that the neural network model can represent rate constants very well.In order to test the prediction ability,for each reaction,the remaining（n-1）groups of reaction data are used as the training set to train the neural network to predict the reaction rate constant.Most of the prediction results are better than the transition state theory.Chapter 4 studies the prediction of rate constants for the reactions of alkanes and oxygen atoms.Using the 11 reactions of reaction rate constants of alkanes and oxygen atoms,a neural network model was established,and the rate constants of 11 sets of reactions were predicted through the remaining（n-1）sets of data,and came up with five schemes to optimize the results predicted by this model.The accuracy of model predictions was improved.Chapter 5 studies the prediction of rate constants for the reactions of alkanes and hydroxyl radicals.Through training 11 reaction rate constants,a multi-layer neural network model is established,and 5 schemes are designed to improve the prediction accuracy.In addition,this work also predicts the reaction rate constants of a total of 664 isomers of alkanes with no more than 12 carbons and hydroxyl radicals at different reaction sites.The results show that the neural network model proposed in this paper has good robustness in predicting specific sites and overall rate constants.Chapter 6 summarizes the full text and looks forward to the future work.