A STUDY OF MODEL REDUCTION BY CONTINUED-FRACTION AND DOMINANT-MODE APPROACHES

Abstract

The importance and relevance of model order reduction can be verified from the huge number of works published regarding the subject. For realization, control, design, computation and other purposes, it is often desirable adequately to represent a high order system by a low order model without sacrificing the important characteristics of original system. Model reduction methods are classified into three categories according to their underlying principles. The three categories are continued fraction expansion, dominant mode, and optimum fitting. The first category of methods, continued fraction expansion, is based on obtaining a reduced model which matches some time moments and Markov parameters of the original model [1]. The second category of methods, dominant mode, is to preserve the dominant elgen values of the original system [10] and the selection of these dominant eigenvalues is made either taking residue at respective elgen value or by defining a weighting factor diagonal matrix. The third category, optimum fitting, is based on identification techniques to obtain a reduced model matching either the time-domain response or the frequency domain response of the original model [15]. But much of stress is laid on first two categories in this dissertation. Given numerical examples, the effectiveness and the merit of simplified models are evaluated on the basis of the simulation of the step response and frequency response. Flow-charts of the software are added in the appendices I-IV.