A novel approach to linear multivariable model reduction is presented based on the minimization of an error function defined in the frequency domain. The new approach is a generalization of Kiendl's method for simplifying single-input single-output systems by means of complex-curve fitting to multivariable systems. Starting with the state-space description of the original model, the reduced-order model is evaluated analytically by minimizing a squared error function at various values of frequency. The new method ensures a design of reduced-order models in dependence on the user's specifications. This is possible by means of adjustable parameters, whose influence on the reduced-order model is obvious.
展开▼
