Optimal reduction of linear systems

The use of direct search (LJ) optimization procedure in multi-pass manner i n the frequency domain is shown to be a viable approach for obtaining the o ptimal reduced model of a linear single-input single-output (SISO) system, for which the optimal solution has eluded previous investigators. The resul ting second-order reduced model provides very close time-domain response of the original model to both impulse and step inputs. Establishment of the t hird-order reduced model is computationally more demanding. Similar behavio ur is observed with an eigth-order transfer function, where the second-orde r reduced model is very easily obtained, but higher order reduced models re quire a considerably larger number of random points to be used in each iter ation. (C) 1999 The Franklin Institute. Published by Elsevier Science Ltd.