Work

Editorial

Year: 2009
Type: editorial
Abstract: The emergence of subspace-based identification methods more than a decade ago has stimulated a lot of research and interest in the development, understanding and use of the new algorithms for the identification of both linear and nonlinear systems. The central goal of the first generation of subspace identification methods was the estimation or better the low-rank approximation of the column or row space of a matrix derived by linear projection from structured (block-Hankel) matrices filled with measurements taken from the system to be identified. This low-rank approximation is the key in deriving approximations of the system matrices of a state–space model. The first family of subspace identification methods has their roots in the Ho–Kalman realization theory, presented almost half a century ago. Their formulation based on reliable linear algebra methods, such as the QR factorization and singular value decomposition, made them on the one hand very attractive for practicioners tackling real-life, multivariable identification problems of high-order dynamical systems. On the other hand, the subspace-based methods represented disruptive changes for the system identification community as the well-established framework of analyzing the accuracy of the estimates via covariances of the estimated parameters, etc. could not straightforwardly be applied to the methods. The integration of subspace methods in high-quality software packages and the growing understanding of the statistical properties, the modifications necessary to use subspace methods in a closed-loop operation, the extension to various classes of nonlinear identification problems, the derivation of models using frequency domain data or the identification of continuous-time models and many more elements have given subspace-based identification a well-established role in the field of system identification. With this special focus section, the goal is to see and stimulate the relevance this evolution in system identification may have on other areas in systems and control. Interesting links are with the areas (1) (adaptive) control; (2) parametric system identification; and (3) signal processing. In this issue the emphasis is restricted to exploring the link with the first two adjacent areas. Within adaptive control the availability of recursive implementations that operate in an online manner can be crucial. This recursive feature may also help in online condition monitoring. By using subspace-based identification in a control context, the tuning of the parameters of the methods is still an open research area. Finally, the use of the combination of subspace-based methods in relation with parametric model identification raises the research question on which model parameterization to use. This special issue collects two papers that explore the above potential links. The first paper authored by Harish J. Palanthandalam-Madapusi and Dennis S. Bernstein. This paper considers the concept of input and state observability, that is, conditions under which both the unknown input and initial state of a known model can be determined from output measurements, providing necessary and sufficient conditions for input and state observability in discrete-time systems. Next, a subspace-based identification algorithm that identifies the state–space matrices and reconstructs the unknown input using output measurements and known inputs is developed. Finally, several illustrative examples are presented, including a nonlinear system in which the unknown input is due to the endogenous nonlinearity. The second paper authored by Graciela González considers adaptive control design for a class of SISO nonlinear discrete-time systems in parametric-strict-feedback form. No growth restrictions are assumed on the nonlinearities. The control objective is to achieve tracking of a reference signal. As usual, the algorithm derives from the combination of a control law and a parameter estimator (certainty equivalence principle). To overcome certain drawbacks of previous schemes, some ideas are introduced: (a) instead of the dead-beat control, a control law is developed, which depends on a set of design parameters contributing to the reduction of control effort; (b) the classical Gram-Schmidt orthogonalization process used for the parameter estimator is replaced by a modified version just to preserve numerical stability; and (c) these novelties together with the calculus of future states in advance yield to an alternative control scheme. When applied to the proposed class of systems, global boundedness and convergence are achieved. As an associate editor, I would like to thank Prof. M. Grimble and the Wiley team for their cooperation.
Institution: Lund University
Language: English
Open Access status: bronze
OpenAlex ID: W1810155571
Cites:
Cited by: 2
Related to: 10