Output-error system identification in the presence of structural disturbances

O2 Other

Internal Authors/Editors

Publication Details

List of Authors: Amir H. Shirdel, Jari Böling, Hannu T. Toivonen
Editors: S. Skogestad
Publisher: NTNU
Place: Trondheim
Publication year: 2015
Book title: Proceedings of the 19th Nordic Process Control Workshop : on a coastal steamer from Trondheim to Bodø


In experimental system identi cation, disturbances can affect destructively estimation of the system parameters. Therefore, finding and eliminating these disturbances can increase signi cantly the accuracy of system identi cation. In our approach, by using orthogonal Laguerre or Kautz basis functions along with sparse optimization and and l1-relaxation, the basis function parameters and structural disturbances are identified simultaneously. Important properties of using Laguerre or Kautz basis functions are to cover system delays and reduce the model complexity compared to finite impulse response models. The output error identi cation models are more reliable and robust for multi{step prediction than other model structures. Using nonlinear kernel methods, such as radial basis functions, some nonlinear cases, like Wiener model, can be identi ed. In addition, sparse optimization with l1-regularization can give simpler solutions and simpler kernel models. In this contribution we present the method and demonstrate it on simulated linear and nonlinear examples, and on real distillation column data. The results show that with the proposed method accurate system models can be identified using experimental data containing unknown trends and outliers.

Last updated on 2020-26-05 at 06:29