Keywords:
Nonlinear systems, Wiener-Hammerstein models, Parameter estimation, Output error, Least squares, Prediction error, Instrumental variables, Genetic algorithms.

Abstract:
The model description for dynamic linear systems is relatively straightforward. Modeling nonlinear processes, however, is more involved due to the lack of a common structure for such systems. It has been shown in the literature that Wiener-Hammerstein models (also known as block cascaded models) can adequately represent a wide range of nonlinear systems. A Wiener-Hammerstein model is composed of a dynamic linear system in cascade with a static nonlinear element followed by another dynamic linear system.

It is the purpose of this research to investigate extending the classical offline and online parameter estimation algorithms to estimate parameters in Wiener-Hammerstein models. In addition, Genetic Algorithms are to be applied for this estimation problem.

The talk starts by addressing the merits of using Wiener-Hammerstein models in representing nonlinear systems. The parameter estimation problem of interest is then formulated. Output Error, Least Squares, Prediction Error, and Instrumental Variable methods are derived both offline and online. Also Genetic Algorithms are applied to estimate the parameters of the above models. Extensive simulation results are presented in order to study and analyze the properties of these estimators. Finally, conclusions and recommendations for future work are highlighted.

Presentation Slides:
Parameter Estimation from Wiener-Hammerstein Models [MS PowerPoint format]