Statistical degree screening method for combustion mechanism reduction

摘要

From a statistical perspective, large complex reaction networks have the power-law degree distributions,which means a small fraction of hub nodes are responsible for most of the network function, while theremaining large amount of low degree nodes give negligible contributions. Based on this, a statisticaldegree screening (SDS) combustion mechanism reduction method was proposed and verified in 42 popularmechanisms of different size and developed by various institutes. It is confirmed that most largecombustion networks obey the power-law form degree distribution with limited fitting error. Then a reducedmodel could be obtained by setting a degree threshold according to the degree distribution andprediction error requirements to distinguish the important species to keep. Intensive validation results inthe comparison with the directed relation graph (DRG) method-reduced mechanisms prove that despitethe SDS method requires considerably less input information and processing time, it can come up withsignificantly reduced models with comparable or even better prediction ability over a broad parameterrange. The good reduction application results of SDS shown here indicates a brand-new angle for largecombustion mechanism analysis and reduction, i.e., from the statistical network structure properties aspect.It is of practical significance for the swift analysis and reduction of large combustion mechanismsand can considerably avoid the influence coming from the dynamical parameters that introduce greatuncertainty.