Comparing different approaches to model error modeling in robust identification (Q1614297)

This paper gives three different approaches to robust identification: stochastic embedding, model error modeling and set membership identification. Each of these methods explicitly accepts the presence of bias model errors, which may motivate the term ``robust identification''. The first two approaches have been developed in the statistical framework, while the latter relies on UBB error assumptions. The nominal model is obtained via least squares estimation from frequency domain data. The uncertainty associated with the model is evaluated from statistical properties of the random walk process describing the unmodeled dynamics. A discussion of asymptotic properties of all methods is presented. An example (where a nontrivial undermodeling is ensured) is given to compare these methods.