For reduced order modelling of large scale second order systems while preserving the second order structure, we propose two approaches based on matching some of the moments and/or Markov parameters of the transfer functions of the original and reduced systems. The first method is based on applying a projection directly to the original system using the so called Second Order Krylov Subspace where the projection matrices are calculated using an extension of the Arnoldi or Lanczos algorithms. By reducing to order Q, this method matches at most Q characteristic parameters, which is less than the standard Krylov methods. In the second method, the number of matching parameters is increased to 2Q in a cost of more computational effort. This approach is based on reduction of an equivalent state space model by a Krylov subspace method matching the first Markov parameter and back conversion into a second order representation by applying a similarity transformation.
«
For reduced order modelling of large scale second order systems while preserving the second order structure, we propose two approaches based on matching some of the moments and/or Markov parameters of the transfer functions of the original and reduced systems. The first method is based on applying a projection directly to the original system using the so called Second Order Krylov Subspace where the projection matrices are calculated using an extension of the Arnoldi or Lanczos algorithms. By re...
»