Model reduction of discrete systems using the power decomposition method and the system identification method

A mixed method of discrete system model reduction retaining the advantages of the power decomposition method and the system identification method is proposed. From the viewpoint of energy contributions to the system output, the dynamic modes with dominant energy contributions, instead of those with dominant eigenvalues, will be preserved by power decomposition method. Having determined the denominator of the reduced model, the parameters of the numerator are found by system identification technique. The reduction procedure is fully computer-oriented. The reduced model is always stable if the original one is stable. Moreover, the reduced model gives good approximation in both the transient and the steady-state responses of the original system.