MODEL ORDER REDUCTION USING MAXIMAL REAL PART NORMS

Prof. Alexandre Megretski, (MIT)

Finding simple and accurate models of complex dynamical systems is a major challenge. For stable LTI systems, when complexity is understood as system order, and accuracy is measured by the H-Infinity norm, the classical method of Hankel model order reduction is known to produce reduced models of guaranteed relative suboptimality. However, there are several disadvantages of Hankel model order reduction:

(a) a state-space model of the original system must be available;

(b) complexity of the calculations grows quickly with the original number of states;

(c) the relative suboptimality guarantees do not extend to approximations with respect to weighted H-Infinity norms.

In the talk, an alternative approach to model order reduction, based on replacing the Hankel norm by the maximal real part norm, will be discussed. The new method apparently does not have the drawbacks (a)-(c). Theoretical statements as well as results of numerical experiments will be presented.

Back to Control Seminars Page.