Premium
Order and parameter estimation of time‐varying system by subspace method
Author(s) -
Tamaoki Morimichi,
Akizuki Kageo,
Oura Kunihiko
Publication year - 2006
Publication title -
electrical engineering in japan
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.136
H-Index - 28
eISSN - 1520-6416
pISSN - 0424-7760
DOI - 10.1002/eej.20169
Subject(s) - subspace topology , singular value decomposition , mimo , algorithm , factorization , computer science , identification (biology) , system identification , qr decomposition , stability (learning theory) , matrix decomposition , mathematics , control theory (sociology) , eigenvalues and eigenvectors , artificial intelligence , data mining , machine learning , control (management) , computer network , channel (broadcasting) , botany , physics , quantum mechanics , biology , measure (data warehouse)
This paper proposes an identification algorithm for time‐varying systems. We apply subspace method for estimation, since it is known to be useful when the input–output (I/O) data are observed by multi‐input multi‐output (MIMO) systems. Among many proposed techniques of subspace methods, we use MOESP (MIMO Output‐Error State Space model identification) in this paper, which assures arithmetic stability by RQ factorization and singular value decomposition (SVD). Generally, subspace methods can be applied after I/O data collection, so that we introduce updated steps of matrices for PI‐MOESP, which uses past inputs for instrumental variables. We propose a recursive update algorithm of PI‐MOESP, including estimation step of the system order, and consider some parameters inherent to the algorithm, namely, initial number of data, estimation step of the order, and forgetting factor. A numerical example shows the usefulness of the proposed method. © 2006 Wiley Periodicals, Inc. Electr Eng Jpn, 157(2): 57–64, 2006; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20169