PARAMETER-IDENTIFICATION FOR UNCERTAIN PLANTS USING H-INFINITY METHODS

We demonstrate the effective use of H-infinity filtering and cost-to-c ome methods for parameter identification in (deterministic) uncertain plants that are linear in the unknown parameters, but nonlinear otherw ise. The cost-to-come method is an approach that has been used earlier to solve linear and nonlinear H-infinity optimal control and filterin g problems. It consists of constructing a cost-re-come function, which assists in the design of an 'optimal' observer scheme. The method is used here in the design of a parameter identification scheme for uncer tain plants, where measurements on the state of the system are availab le, but not on its derivative. Two approaches are adopted, in both of which the parameter estimation problem is formulated as an H-infinity filtering problem. One of the approaches uses a more standard prefilte ring of the past states, input and disturbance signals. The other appr oach is a novel design method, which leads to a new class of identific ation schemes. It involves two subproblems: FSDI (full-state-derivativ e information) problem, where it is assumed that both the state and it s derivative are available to the parameter estimator, and NPFSI (nois e-perturbed FSI) problem, where the estimator is assumed to measure a noise-perturbed measurement of the state. In the latter problem we use singular perturbation methods to prove asymptotic convergence of the performance of the identifier to that of the unperturbed case, thus pr oviding an asymptotically optimal solution to the FSI (full-state meas urement) problem. To illustrate both approaches, several simulation st udies on a numerical example are provided.