Structured total least norm for MISO ARX system identification

• Autorzy: Larkowski T. , Vinsonneau B. , Burnham K. J.
• Języki publikacji: EN

Abstrakty

EN

The errors-in-variables (EIV) identification framework concerns the identification of dynamic models of systems where all the variables are corrupted by noise. The total least squares (TLS) is one of the most prominent techniques that has proven to be both robust and reliable. The structured total least norm (STLN) can be seen as a natural extension to TLS that preserves any affine structure of the joint data matrix, which is mostly the case in identification schemes. In contrast to the least squares (LS), TLS or mixed LS-TLS problems, the STLN solution cannot be expressed in a closed form, therefore, an optimization procedure is required. Note that STLN allows different norms to be considered other than the usual square norm (or 2 norm). This paper describes a direct application of the STLN approach for systems that can be represented by auto-regressive with exogenous input (ARX) multi-input single-output (MISO) models. The performance of the proposed STLN algorithm (in the case of the square norm) is compared to the LS, the bias-eliminating LS (BELS), the extended matrix LS (EMLS), the instrumental variables (IV), TLS and the compensated TLS (CTLS) methods when applied to a simulated MISO ARX system. Results, obtained from Monte Carlo simulation, show that, under the conditions considered here, STLN surpasses all other investigated techniques, attaining the best estimates of the true system parameters.

Słowa kluczowe

EN

errors-in-variables; least squares; structured total least norm; system identification

• Wydawca: Oficyna Wydawnicza Politechniki Wrocławskiej
• Czasopismo: Systems Science
• Rocznik: 2007
• Tom: Vol. 33, no 4
• Strony: 73--83
• Opis fizyczny: Bibliogr. 13 poz., wykr.

Twórcy

autor

Larkowski T.

autor

Vinsonneau B.

autor

Burnham K. J.

• Coventry University, Control Theory and Applications Centre, Priory Street, CV1 5FB Coventry, UK,

Bibliografia

• [1] Lemmerling P., Structured Total Least Squares: Analysis, Algorithms and Applications, PhD Thesis, Catholic University of Leuven, 1999.
• [2] Linden J.G., Vinsonneau B., Burnham K.J., Review and comparison of some identification methods in the errors-in-variables framework, Proc. 17th Int. Conf. Systems Engineering, Coventry, UK, 2006, pp. 243-254.
• [3] Mańczak K., Nahorski Z., Komputerowa identyfikacja obiektów dynamicznych, Państwowe Wydawnictwo Naukowe, Warsaw, Poland, 1983.
• [4] Markovsky I., Van Huffel S., Overview of Total Least-Squares Methods, Signal Processing, Elsevier, 2007, pp. 2283-2302.
• [5] Rosen J.B., Park H., Glick J., Total least norm formulation and solution for structured problems, Journal on Matrix Analysis and Applications, SIAM, 1996, pp. 110-126.
• [6] Rosen J.B., Park H., Glick J., Structured total least norm for nonlinear problems, Journal on Matrix Analysis and Applications, SIAM, Vol., 20, 1998, pp. 14-30.
• [7] Söderström T., Soverini U., Mahata K., Perspectives on errors-in-variables estimation for dynamic systems, Signal Processing, Vol. 82(8), 2002, pp. 1139-1154.
• [8] Söderström T., Why are errors-in-variables problems often tricky? Proc. European Control Conference ECSE 2003, Vol. 82(8), 2003, Cambridge, UK.
• [9] Söderström T., Errors-in-Variables Methods in System Identification, Automatica, 2007, pp. 939-958.
• [10] Vandersteen G., On the use of compensated total least squares in system identification, IEEE Trans. Automatic Control, Vol. 43, 1998, pp. 1436-1441.
• [11] Van Huffel S., Vandewalle J., The Total Least Squares Problem: Computational Aspects and Analysis, SIAM, Philadelphia, USA, 1991.
• [12] Van Huffel S., Park H., Rosen J.B., Formulation and solution of structured total least norm problems for parameter estimation, IEEE Trans. on Signal Processing, Vol. 10, 1996, pp. 2464-2474.
• [13] Van Loan C.F., On the method of weighting for equality constrained least squares problems. Numerical Analysis, Vol. 22, 1985, pp. 851-864.