Parameterized Model Order Reduction by Superposition of Locally Reduced Bases

Abstract

We present an approach for model order reduction of parameterized linear dynamical systems which combines local reduced bases using singular value decompositions. The local reduced bases are computed for a fixed set of sample points in the parameter domain by moment matching techniques. Covering the parameter domain with sample points may yield a very large set of samples. We investigate a superposition approach that takes only a few of the local models into account to keep the resulting reduced dimension small. Our approach will be compared to one existing approach which directly interpolates local bases in some illustrating numerical experiments.