Identification and Simulation of Nonlinear Dynamics Using a Dynamical Local Model Network

Christoph Weise and Kai Wulff*

Physical models are often hard to obtain due to complex dynamics and uncertain parameters. In particular in automotive applications high precision physical models are rare. In recent years local model networks (LMN) are used more frequently to describe complex nonlinear dynamics by a number of simple linear models. Here some significant set of variables is chosen and devided into partitions for which a local linear model is identified. In this contribution we consider a turbo-charger system for a diesel engine and apply an LMN approach to identify the input-output dynamics of the system. We are investigating several issues in regard of the selection of the parameter space and interpolation method as well as the application of the local linear model tree (LOLIMOT) approach. The obtained LMN is compared to experimental data of an independent test cycle.

Mathematics Subject Classification: 93A30

Keywords: mathematical modelling, system identification, local model network

Minisymposion: Modelling and Control of Mechatronic Systems