Machine Learning Research On Rate Constants Of Reactions Between Alkane And Free Radical

The reaction rate constant is an important parameter in the study of chemical kinetics,and it is one of the main research objects of experimental and theoretical chemistry.On the one hand,the research on the combustion reaction rate of fuel can reveal its combustion characteristics and improve the fuel combustion efficiency.On the other hand,it can effectively control the emission of environmental pollutants.Since experimental measurements and theoretical calculations of reaction rate constants are often expensive,only a few rate constants have been obtained.In recent years,it has been shown that machine learning can provide some new ideas for predicting chemical reaction rate constants.In this study,different machine learning algorithms were used to build reaction constant prediction models to predict the rate constants of the reactions between alkane and CH3/HO2/H/O radicals.In addition,a hybrid machine learning model was developed,combining the advantages of neural network models and decision tree-based algorithms,which can improve the prediction accuracy of the ML models.The full text is divided into 6 chapters,the main contents are as follows:Chapter 1 introduces the background of the scientific problem studied in this paper,the current research progress at home and abroad,and related research methods.Chapter 2 takes the reaction rate constants of alkanes and CH3 radical as the research object,14 reactions were collected and the corresponding data set was constructed which includes a total of 3075 data points.The descriptors were calculated and filtered,and then 6 descriptors were screened.Different machine learning models were constructed and a new XGB-FNN hybrid model was proposed.The model evaluation results showed that the hybrid model has the best comprehensive performance.The rate constants for the primary hydrogen abstraction reactions of nalkanes with 5 to 12 carbon atoms in the temperature range from 300 to 2500 K were predicted,and the results showed that the predicted rate constants follow the threeparameter Arrhenius equation well,which are in good agreement with the experimental measured or theoretically calculated rate constants.Chapter 3 takes the reaction rate constants of alkanes and HO2 radical as the research object.A total of 10 reactions were collected,and a corresponding data set was constructed,which contains a total of 2313 data points.Three different machine learning models were built using the method of Chapter 2.The model evaluation results showed that the constructed models have good predictive accuracy and stability.Chapter 4 takes the reaction rate constants of alkanes and H atom as the research object.By using the method of Chapter 2,a total of 15 reactions are collected,and a corresponding data set is constructed,which contains a total of 3705 data points.Three different machine learning models were built using the method of Chapter 2.The model evaluation results showed that the constructed models have good predictive accuracy and stability.The rate constants of three reactions in the temperature range from 300 to 2500 K were predicted,and the results showed that the predicted rate constants follow the three-parameter Arrhenius equation well,which are in good agreement with the experimental measured or theoretically calculated rate constants.Chapter 5 takes the reaction rate constants of alkanes and O atom as the research object.A total of 22 reactions were collected,and a corresponding data set was constructed,which contains a total of 7502 data points.Descriptors were calculated and 6 descriptors were screened,and several different machine learning models were constructed.The model evaluation results showed that the constructed models have good accuracy and stability.Chapter 6 summarizes the full text and gives an outlook.