A COMPARISON OF DESCENT AND CONTINUATION ALGORITHMS FOR H-2 OPTIMAL, REDUCED-ORDER CONTROL DESIGNS

Reduced-order control is important in control engineering practice due to inevitable limitations on the throughput of the control processors . This paper considers the direct design of H-2 optimal, reduced-order controllers via parameter optimization approaches. Two classes of num erical methods are considered: (1) descent methods, and (2) continuati on (or homotopy) methods. In particular, three standard descent method s (i.e. steepest descent, conjugate gradient and BFGS Quasi-Newton) ar e compared with a recently developed continuation algorithm by using t hree examples appearing the literature. The results indicate that the BFGS algorithm tends to be more efficient than the continuation algori thm, which in turn tends to be more efficient than the two remaining d escent algorithms. An important observation is that each of the algori thms appears to be most efficient when the controller is constrained t o a tridiagonal basis.