ABSTRACT
This chapter considers how one's approach to modelling changes when one has a particular control law in mind. It provides an overview of the multi-model approach to modelling, as this gives some synergy with Model Predictive Control. The chapter introduces the principle concept of utilising the multi-model but does not focus on more advanced details of the identification algorithm. Model identification could be considered as an end in itself; however, often the model is wanted as a base for control law design and hence the potential utility of the model for this purpose must be considered during the identification. The multi-model is far less sensitive to parameterisation errors than prediction models based on one-step ahead models. The multi-model approach gives better accuracy when the order is underparameterised and when overparameterised, does not suffer from the near cancelling pole/zero effects that predictions based on one-step ahead models suffer from.
