Model order reduction by state vector selection (SVS) approach

A new approach to construct multipoint projection-based model order reduction algorithms is proposed. The approach is aimed to decrease the redundancy of the reduced model and to provide an effective error control. The generation of the projective matrix is performed using worst-case analysis to determine frequency point, input excitation and internal state vector corresponding to the maximal value of the user-defined error norm. The value of the norm also provides the error control and stopping criterion. Numerical results of the comparison of the proposed approach with known methods PRIMA and PMTBR are presented.