Parameter identification for uncertain plants using \(H^ \infty\) methods (Q1899555)

The use of \(H^\infty\) filtering and cost-to-come methods are applied to worst case parameter identification in deterministic but uncertain plants that are linear in the unknown parameters. The approach consists in constructing a cost-to-come function which assists in the design of an ``optimal'' observer. Full state measurements are assumed. One variant uses pre-filtering of past states, inputs and disturbances to estimate the derivative information. Another solves two subproblems: Parameter identification based on the assumption of full state with derivative information and use of an estimator of noise perturbed state measurements. It is shown using singular perturbation methods that the identifier using noise perturbed measurements asymptotically converges to the unperturbed case. In this sense, the full state measurement without derivative information parameter identification problem is solved in an asymptotically optimal way. A numerical example is included.