Stable reduced Order Modelling of Large Scale Systems using Prescribed Poles
Abstract
By parameterizing all reduced order models matching some of the first characteristic parameters (moments or Markov parameters or both) of the original system, we prescribe the poles of the reduced system as a key to find a stable reduced order system. Two procedures based on Lanczos and Arnoldi algorithms are proposed to approximate the dominant poles of the original system and to choose them as the poles of the reduced order model. If the number of iterations approximating the dominant poles is more than the order of the reduced order model, the new approach leads to a better result than thestandard Krylov subspace methods.