On robustness in system identification (Q1298235)

This paper studies robustness issues in system identification. A framework for robust convergence analysis is given. It uses a set of results on robust convergence under average \(l_p\) norms, Orlicz norms and cross-correlation-type noise semi-norms. The new results are used to discuss and to study robustness issues in identification of linear time-invariant models when the true plant is not exactly linear. The results show that the effects of unmodelled dynamics are typically more serious than those of random noise. A divergence result is also given and additive error analysis is presented.