HyChem method based on BP neural network optimization

A kinetic model study on aviation kerosene （JetA-2） is conducted by combining Hybrid Chemistry （HyChem） and BP neural network prediction methods. A multi-objective genetic algorithm （NSGA-II） is implemented to develop a fifteen-component surrogate model for JetA-2， incorporating physicochemical properties to obtain thermodynamic data and predict ignition delay times under specified conditions. The BP neural network prediction model is utilized to determine the stoichiometric coefficients and reaction rate constants for the seven-step lumped reaction mechanism， resulting in the aviation kerosene （JetA-2） HyChem reaction kinetics model， which includes 113 species and 791 elementary reactions. Comparing the ignition delay times and laminar flame speeds of the developed HyChem reaction kinetic model with those developed using traditional method， functional groups for Mechanism （FGM）， and stochastic gradient descent （SGD） method， the results indicate that the developed HyChem reaction kinetic model presented in this paper exhibits high predictive accuracy， with a relative error in ignition delay time as low as 12.7% and a relative error in laminar flame speed as low as 1.8%， both of which are superior to those of HyChem reaction kinetic models using other methods.