Financing: Research Foundation - Flanders (FWO)

Project reference Nr.: G.0377.12
Start: 2012-01-01
End: 2015-12-31

Description:

The aim of this project “Modelling structured dynamical systems: parametric and non-parametric approaches” is to realize synergies between different families of modelling and identification techniques for structured nonlinear dynamical systems. A first class of methodologies relates to support vector machines, originally developed within the machine learning community. A second class is based on classical system identification, which finds its roots mainly within the systems and control community. Parametric and non-parametric approaches will be studied from these different perspectives. This project aims at achieving complementary and unifying insights with:- a better understanding of the impact of the intrinsic assumptions, e.g. how crucial is it to impose certain models structures in predictive modelling?- a better understanding of the strong and weak points of the different approaches: why is one approach successful on certain types of problems while it may totally fail on others?- proposing new methods that combine and integrate the best of different modelling paradigms- data driven detection of the dynamic structure of the nonlinear system

SMC people involved in the project:

• Johan Schoukens (Co-promoter)
• Johan Suykens (Promoter)