Over the last years the importance of the reduction of combustion engine vehicle emissions has steadily increased. Thus today the calibration of combustion engines focuses to an even greater extent on the reduction of unwanted emissions. For this complex task the use of simulation models has been very successful. In many cases steady state models are used. However, their capability to predict emissions in dynamic scenarios is very limited. As a consequence, dynamic emission models are gradually coming into the spotlight. These models require more and more sophisticated identification methods for highly nonlinear dynamic systems. In this study multiple approaches to dynamic modelling of nonlinear dynamic systems have been compared on carbon hydride (HC) and the nitric oxide (NOX) emissions. Models from the System Identification Toolbox from Mathworks have been compared to approaches which were likewise implemented object-oriented in MATLAB or simply in a script. The results have shown several promising approaches from each of the implementations. As all model types tended to overfitting on different parts of the training data it seems only reasonable to familiarize oneself with multiple of the approaches to switch between the models if necessary. Especially highly flexible model structures with fast trainings algorithms like the Volterra Series, the Hammerstein-Wiener models or the recurrent Wavenets are useful, as a high number of parameter constellations/model structures can be tested in a short time.
Comparison of identification methods for nonlinear dynamic systems
2011
20 Seiten, 10 Bilder, 2 Tabellen, 21 Quellen
Aufsatz (Konferenz)
Englisch
A parametric identification technique for nonlinear dynamic systems
British Library Conference Proceedings | 1994
|