Application of Surrogate Modelling to the Optimisation of Kinetic Parameters in an Emissions Control Catalyst Model Using Vehicle Drive Cycle Data

Abstract

This work describes the application of surrogate modelling to the optimisation of kinetic parameters in a kinetic model of a vehicle emissions control catalyst using engine or vehicle test data. Optimisation using conventional methods with a detailed kinetic model is slow as the model requires a relatively long time to simulate each drive cycle. The use of a surrogate model as a substitute for the detailed model can considerably accelerate the optimisation by drastically reducing the number of evaluations of the detailed model required. A surrogate model is simply a function that approximates the behaviour of the full model; accuracy is sacrificed for speed of evaluation. A few runs of the detailed model are required to train the initial surrogate. The parameters for these are determined using Latin hypercube sampling. After that, the surrogate model is used in place of the detailed model with a simplex optimisation method. Since evaluation of the surrogate is computationally inexpensive, the optimisation runs rapidly. The kinetic parameters obtained can then be evaluated with the full model and the surrogate model updated with the new information. Thus, the surrogate model is improved with each iteration. As an example of application, the extension of an existing kinetic model for a diesel oxidation catalyst to a model capable of predicting the effect of platinum group metal (PGM) loading (30 to 120 g ft.−3) using drive cycle data is presented. This optimisation methodology is useful for extending existing kinetic models to different agings, PGM loadings, catalyst formulation, etc.