Structural Identification and Parametric Estimation of Multivariable Systems

Abstract

<p>The identification of discrete linear time-invariant multivariable systems has been investigated. This thesis consists of three interrelated parts. In the first part, linear systems have been described in canonical forms which are mostly suitable for system identification. Inter-relationships between the input-output difference canonical form due to Bonivento, Guidorzi and Marro (1973) and the state-space canonical form due to Luenberger (1967) have been studied. In the second part structures of linear multivariable systems have been studied. An efficient algorithm has been developed to identify a set of structural indices directly from the input-output data of a system. The order of the system is, then, determined in terms of these structural indices. In the last part, estimation of system parameters from the input-output data has been studied. Recursive algorithms have been developed for estimating the system parameters sequentially from noise-free and noise-corrupted data. Digital computer simulations with noise-free as well as noise-corrupted data have been utilized to illustrate the usefulness of the proposed procedures.</p>