New approach for microkinetic mean-field modelling using latent variables

Abstract

A detailed microkinetic model for the storage and reduction steps in a lab-scale NO X storage reactor was taken from the literature and a large number of parameters were subject to fitting to a new set of experimental transient data. The parameters chosen for fitting were selected by sensitivity analysis of the Jacobian matrix of all adjustable parameters using latent variable (LV) models (partial least squares, PLS). The analysis of the LV structure of the Jacobian matrix is important because it indicates the experimental rank, i.e. how many parameters that are relevant to fit. Two slightly different methods both using LV models are presented in order to select good candidate parameters for fitting as well as a method to fit linear combinations of many parameters that only span the experimental space. By using this methodology the risk of overfitting is reduced and the chances for successful fitting are increased by supplying an appropriate number of parameters that also are uncorrelated (independent) for the given experiment. This methodology contributes to a better understanding of the catalytic process as well as a potential for more efficient parameter fitting for complex transient heterogeneous catalytic systems. The application of this kind of sensitivity analysis offers a new approach to microkinetic modelling.