Communications in Information and Systems
Volume 19 (2019)
Machine learning of the rate constants for the reaction between alkanes and hydrogen/oxygen atom
Pages: 391 – 403
The reaction rate constant is critical in modeling combustion and biochemistry reaction network. In this work, a machine learning approach using multi-layered neural network (NN) models is applied to training and predicting rate constants of combustions reactions. Two kinds of hydrogen abstraction reactions are considered: Hydrogen + Alkanes (HR) and Oxygen + Alkanes (OR). Each reaction is described by five parameters: three of which distinguish the structure of the reactant alkane, one is the serial number of the carbon atom for the broken C-H bond and the last one is the temperature. Two NN models are trained separately by fitting the rate constants of eight HR or eleven OR reactions. The small deviations indicate that the rate constants can be well represented by the NN models. To test the predictive ability, one model is constructed for each reaction by fitting the rates constants of the rest $n-1$ reactions ($n = 8$ for HR reactions and $n = 11$ for OR reactions). The deviations are 25.3%-2396.3% for the HR reactions and 15.0%-659.4% for the OR reactions. Most of the prediction results are better than those from the transition state theory. Overall, the machine learning approach is an efficient method to predict chemical reaction rate constants.
machine learning, neural network, rate constant, alkane
Received 30 September 2019
Published 15 April 2020