Einladung zum Vortrag im Kolloquium Technische Kybernetik

Closed-loop Subspace Identification: A New Orthogonal Projection Approach

Prof. Biao Huang, Ph.D. P.Eng
Department of Chemical and Materials Engineering
University of Alberta, Canada

    Zeit: Dienstag · 01. 06. 2004 · 16:00 Uhr
    Ort: Raum V 9.31 · Pfaffenwaldring 9 · Campus Stuttgart-Vaihingen

Abstract

The classical closed-loop identification methods such as prediction error method have been developed over the last 30 years. Important issues such as identifiability under closed-loop conditions have received attention from many Researchers. A number of identification strategies have been developed successfully. However, these classical closed-loop system identification methods have limitations in terms of complexity in the computation and no guarantee of global convergence. On the other hand, the subspace identification method is a relatively new approach used for the state space model identification and enjoys the properties of simplicity and free of convergence problem. The most representative subspace identification algorithms are N4SID, CVA and MOESP. However, all these algorithms apply to open-loop data only although some modifications have also been made to satisfy closed-loop identification to some extent. A somehow “mysterious” bias error under closed-loop conditions exists in many subspace algorithms. In this presentation, a new closed-loop subspace identification approach through an orthogonal projection and subsequent singular value decomposition is proposed. It explains why some existing subspace algorithms may deliver a bias in the presence of the feedback control and suggests a remedy to eliminate the bias. Furthermore, as the proposed method is a projection based method, it can simultaneously provide extended observability matrix, lower triangular block-Toeplitz matrix, and Kalman filtered state sequences. Therefore, using this method, the system state space matrices can be recovered either from the extended observability matrix/the block-Toeplitz matrix or from the Kalman filter state sequences. Simulations based on a benchmark problem compare the performance of the proposed algorithms with a number of well-known open-loop as well as closed-loop subspace identification algorithms and verify the feasibility and closed-loop applicability of the proposed algorithms.

Biographical Information

1979-1986:B.Sc and M.Sc in Automatic Control, Beijing University of Aeronautics and Astronautics.
1986-1992: Lecturer, Beijing Institute of Technology
1993-1997: Ph.D. in Process Control, University of Alberta, Canada
First half year of 1997: Consultant, Matrikon Inc., Canada
1997- present: Assistant Professor, Associate Professor, and Full Professor, Department of Chemical and Materials Engineering, University of Alberta, Canada
2003-2004: received Alexander von Humboldt research fellowship and currently visiting University of Duisburg.
1997-2003: Recipient of Petro-Canada Young Innovator Award and Canadian Foundation for Innovation Researcher award.
Since 1997, published 57 papers in international journals and 52 presentations in international conferences