A New Machine Learning Algorithm to Optimize A Reduced Mechanism of 2-Butanone and the Comparison with Other Algorithms
Abstract
2-butanone (methyl ethyl ketone) has been identified as a potential alternative fuel and fuel tracer in recent studies. In this work, a reduced mechanism containing 50 species and 190 reactions for 2-butanone is developed for the first time. The raw reduced mechanism is built in three parts using decoupling methodology, a reduced C4-Cn sub-mechanism, a reduced C2-C3 sub-mechanism and a detailed H2/CO/C1 sub-mechanism. Subsequently, the self-adaptive differential evolution algorithm of machine learning is proposed for optimizing the reaction rates of 31 reactions in the C4-Cn sub-mechanism to predict the ignition delay times and laminar flame speeds in constant volume bombs. The optimized reduced mechanism is validated by the ignition delay times in shock tubes and laminar flame speeds in constant volume bombs. The results of the optimized reduced mechanism are similar to those of the detailed mechanism, which show it is reliable. Moreover, the performance of the self-adaptive differential evolution algorithm is much better than the genetic algorithm and the particle swarm optimization.