SVD-Krylov based model reduction for time-varying periodic descriptor systems

In this paper, we propose a projection based iterative technique for model order reduction (MOR) of time-varying periodic descriptor systems in continuous setting. This is a two sided projection technique which combines the SVD and the Krylov subspace method. We first apply a simple descretization scheme on our original periodic model, and reformulate an equivalent time-invariant form. We then reduce it with a two sided projection technique, where one side of this reduction technique is singular value decomposition (SVD) based Gramian approximation, and the other side is based on Krylov based projection. For the Krylov based projection method we mainly focus on the multipoint moment matching technique. The result is that the reduce model is asymptotically stable. Numerical results on a real life model have successfully substantiated the efficiency of our reduction process ensuring the stability of the reduce system.