Model Reduction Over a Frequency Interval Using the Spectral Function

Two simple methods of model reduction over a frequency interval are proposed. In both of the procedures the reduced model is chosen so that its spectral amplitude function matches with that of the original system in a specified frequency band. In the first method we develop a polynomial approximation to the spectral function of the original system. In the second method we force the spectral function of the reduced order model to coincide with that of the original system at certain chosen frequency points. The frequency responses of the reduced models obtained by both methods match fairly well with that of the original system. The virtue of this method is that it can be applied to systems of which only frequency response data are available and not the transfer function. © 1986.