Reduction of lumped reaction networks based on global sensitivity analysis

Abstract

In case of complex chemical systems in that many reactions occur between high numbers of species, it is often necessary to use some form of lumping in order to obtain a simpler kinetic model whose parameters can actually be determined from the available data. Designing a lumped reaction network, however, involves dealing with great degrees of freedom. One has to define the actual pseudocomponents, the reactions that take place in the system and finally, reliable kinetic parameters have to be identified for these reactions. This often results in high uncertainties regarding these kinetic models as well as in their limited usability in carrying out experimental design or other process engineering tasks. This point of interest is more often not get discussed than it is. The key idea of this paper is that the uncertainty of lumped kinetic models can be significantly diminished by reducing the size of the reaction network while also preserving the ability of the model to describe the observed behavior of the given system. To that end, we applied and evaluated five different global sensitivity analysis methods based on their performance, using the case studies of real plastic waste pyrolysis and vacuum gas oil hydrocracking. We point out that these tools can be effectively used to construct lumped reaction networks with fewer parameters to be estimated with narrower confidence intervals.