On-Line Structure Selection for the Identification of Multivariable Systems

Sammanfattning: An algorithm is described for the selection of mod el structure for identifying multivariable mode Is of"black box" character. The algorithm receives as "input" a mode I in a given parametrization (= mode I structure). It is then tested whether this parametrization is suitable (well conditioned) for identification purposes. If not, a better one is selected and the transformation of the mode! to the new representation is performed.The test is based upon the properties ofthe Hankel matrix ofthe impulse response ofthe mode I. The relation between the conditioning ofthe Hankel matrix and the conditioning of the identification (minimization) is described.This mode! structure selection algorithm can be used as a block in both an iterative, offline identification procedure, and in recursive, on-line identification. It can be used whenever there is some indication that the parametrization is ill-conditioned. It is discussed how the algorithm can be interfaced with an identification procedure.A complete procedure consisting of an off-line identification algorithm and the model structure selection algorithm is described. Results are given when this com bined algorithm is applied to both simulated and real data.