Canonical forms for aggregated models†

It is shown that starting with canonical forms of linear multivariable systems, it is possible to obtain an aggregated reduced-order model which retains the same structure as the original canonical form but with smaller blocks. Besides retaining the desired eigenvalues of the original system, the reduced-order model also matches some time moments or Markov parameters. Moreover, the proposed method has many computational advantages over the normal procedures for determining the aggregation matrix.