Subspace identification by data orthogonalization and model decoupling (Q705467)

This paper, like its companion papers [\textit{A. Chiuso} and \textit{G. Picci}, ``On the ill-conditioning of subspace identification with inputs'', Automatica 40, No. 4, 575--589 (2004; Zbl 1168.93330) and ``Numerical conditioning and asymptotic variance of subspace estimates'', ibid. 40, No. 4, 677--683 (2004; Zbl 1168.93411)], deals with ill-conditioning of subspace identification methods of input-output data. It demonstrates that the use of a preliminary orthogonal decomposition of the data, together with appropriate subspace algorithms adapted to this decomposition may, in certain circumstances, lead to more robust and accurate estimates. According to the authors, the possible loss of accuracy in the presence of ill-conditioning is not due to the use of subspace methods. Except for the introduction, this paper consists of the following parts: 2. State-space models for subspace identification. The authors introduce a canonical form with a so-called deterministic and a stochastic subsystem and make the standing assumption that these subsystems do not have common dynamics. In this case they call the overall model decoupled. According to the authors, their main concern is to compare the performance of subspace algorithms applied to the identification of a joint model (that is in this case overparametrized). 3. Identification from data on a finite interval. In this part the authors use a weighted version of the PI-MOESP method of \textit{M. Verhaegen} and \textit{P. Dewilde} [Int. J. Control 56, No. 5, 1187--1210 (1992; Zbl 0772.93020)] for the identification of the deterministic subsystem. 4. Probing inputs for identification. In this section the authors investigate a class of inputs which leads to the largest state-input correlation coefficients and hence enhance the possible ill-conditioning of the identification problem. 5. Some experimental results. 6. Conclusions. The paper ends with an appendix and the list of references. This clearly structured and well-written paper is an important contribution to the large literature about (subspace) identification.