Particle Filters for System Identification of State-Space Models Linear in Either Parameters or States
2003 (English)In: Proceedings of the 13th IFAC Symposium on System Identification, 2003, 1251-1256 vol.1 p.Conference paper (Refereed)
The potential use of the marginalized particle filter for nonlinear system identification is investigated. The particle filter itself offers a general tool for estimating unknown parameters in non-linear models of moderate complexity, and the basic trick is to model the parameters as a random walk (so called roughening noise) with decaying variance. We derive algorithms for systems which are non-linear in either the parameters or the states, but not both generally. In these cases, marginalization applies to the linear part, which firstly significantly widens the scope of the particle filter to more complex systems, and secondly decreases the variance in the linear parameters/states for fixed filter complexity. This second property is illustrated on an example of chaotic model. The particular case of freely parametrized linear state space models, common in subspace identification approaches, is bi-linear in states and parameters, and thus both cases above are satisfied. One can then choose which one to marginalize.
Place, publisher, year, edition, pages
2003. 1251-1256 vol.1 p.
System identification, Nonlinear estimation, Recursive estimation, Particle filters, Kalman filters, Bayesian estimation
Engineering and Technology Control Engineering
IdentifiersURN: urn:nbn:se:liu:diva-13919ISBN: 978-0080437095OAI: oai:DiVA.org:liu-13919DiVA: diva2:22192
13th IFAC Symposium on System Identification, Rotterdam, The Netherlands, August, 2003