Identification des systèmes multivariables : méthodes des sous-espaces. Partie II: applicabilité et intérêt

The aim of this article is to examine the applicability of the subspace methods to system identification. It is based on a comparative study of subspace methods and some « classical » methods from an experimental point of view. Two problems are considered, a simulation issue based on an ill-conditioned heat-exchanger process and in application to a winding pilot plant. Results show the practical aspect of the subspace methods which are able to estimate the multivariable behavior of the two systems by reduced order state-space models while they need only one design parameter: the model order. This study also reveals the limits of the subspace methods as their sub-optimal feature and the lack of information about the uncertainty of the estimated model.