System Identification

Contact Details

Prof. Brett Ninness

Email

Phone

(02) 4921 6032
+61 2 4921 6032 (intl)

Fax

(02) 4921 6993

Office

Callaghan Campus
Building EA: EA-G29

Post

Prof. Brett Ninness

School of Electrical Engineering and Computer Science

- - -

Funding

Australian Research Council

ARC Discovery Project
DP0666955
2006-2008
Value: $336,000

Australian Research Council

ARC Discovery Project
DP0208665
2002-2005
Value: $360,000

Australian Research Council

Discovery Project
DP0774086
2007-2009
Value: $246,090

Australian Research Council

ARC Discovery Project
DP1097142
2010-2012
Value: $330,000

Australian Research Council

Discovery Project
DP140104350
2014-2016
Value: $466,000

Industry Funding

General Motors Partnership
2011-2012
Value: $110,000
Theoretical and empirical study of various problems in system identification. Particular attention is paid to robust estimation of Multivariable and Nonlinear systems, and to error quantification.
Contents

Sub-Projects

MCMC System Identification

Markov Chain Monte-Carlo methods are used to calculate probability density functions for parameters in dynamic systems models. By virtue of computation of the true posterior density, these methods allow accurate quantification of estimation error, even for short data lengths.

System Identification Toolbox

This toolbox is a MATLAB-based software package for the estimation of dynamic systems.

A wide range of standard estimation approaches are supported. These include the use of non-parametric, subspace-based and prediction-error algorithms coupled (in the latter case) with either MIMO state space or MISO polynomial model structures.

Additionally, some new approaches are included. These include the support for bilinear and other Hammerstein-Wiener non-linear structures, and the use of the expectation-maximisation (EM) algorithm for time and frequency domain estimation of state space structures.

Variance Quantification

This project develops quantifications for the frequency domain variance of prediction error system estimates. A key theme is to derive new approximations offering improved accuracy via the principles of reproducing kernel principles and orthonormal parametrizations.
Team Members: Prof. Brett Ninness

Wiener Hammerstein Benchmark

Details of our attempt at the Wiener-Hammerstein Benchmark problem

Maintained by Prof. Brett Ninness
University of Newcastle
23 Jun 2008, © Copyright