Machine learning rate constants of hydrogen abstraction reactions between ester and H atom

Abstract

The rate constants of hydrogen abstraction reactions of alkyl ester by H atom are crucial for optimizing combustion reaction network and improving combustion efficiency of biodiesel. Due to the lack of experimental and theoretical rate coefficients data for some reactions - such as hydrogen abstraction of large biodiesel molecules by free radicals - machine learning provides a viable alternative to predict rate constants. In this study, three different machine learning (ML) methods - feedforward neural network (FNN), extreme gradient boosting (XGB) and XGB-FNN hybrid model - were used to predict rate constants of the reactions between alkyl ester and H atom. The rate constants of 41 reactions between H + CnH2n+1COOCmH2m+1 (n = 0–5, m = 1, 2) were calculated by the Master Equation System Solver (MESS) program over a temperature range of 300–2000 K for model training. The results showed that the XGB-FNN model with 8 descriptors has better overall performance than the other two ML methods. The average deviation of XGB-FNN model on the test set is 33.56% by performing leave one out (LOO) cross validations. The rate constants of the H + methyl decanoate (MD) reactions over a temperature range of 300∼2000 K were predicted by the XGB-FNN model, which follow well the modified three-parameter Arrhenius equation and agree well with theoretical values, indicating that the hybrid XGB-FNN model is robust in predicting the rate constants of alkyl ester and H atom in the temperature range of combustion. The present ML method in this study is supposed to be able to provide accurate and affordable rate constants prediction of larger systems, the molecular sizes of which are comparable to those of the dominant components of real biodiesel. That is important for the development of biodiesel kinetic models.