Optimization of kinetic mechanism for hydrogen combustion based on machine learning

The reduced mechanism based on the minimized reaction network method can effectively solve the rigidity problem in the numerical calculation of turbulent internal combustion engine. The optimization of dynamic parameters of the reduced mechanism is the key to reproduce the experimental data. In this work, the experimental data of ignition delay times and laminar flame speeds were taken as the optimization objectives based on the machine-learning model constructed by radial basis function interpolation method, and pre-exponential factors and activation energies of H₂ combustion mechanism were optimized. Compared with the origin mechanism, the performance of the optimized mechanism was significantly improved. The error of ignition delay times and laminar flame speeds was reduced by 24.3% and 26.8%, respectively, with 25% decrease in total mean error. The optimized mechanism was used to predict the ignition delay times, laminar flame speeds and species concentrations of jet stirred reactor, and the predicted results were in good agreement with experimental results. In addition, the differences of the key reactions of the combustion mechanism under specific working conditions were studied by sensitivity analysis. Therefore, the machine-learning model is a tool with broad application prospects to optimize various combustion mechanisms in a wide range of operating conditions.