Robust adaptive identification of slowly time-varying parameters with bounded disturbances (Q1301444)

Taking into account the limitations of recursive identification, one derives a well-defined direct least-squares estimation algorithm based on the spectral factorization of an information matrix. It is shown that it has better robustness and performance properties than its recursive counterparts at the expense of the on-line computation of an eigenvalue decomposition of a symmetric positive semidefinite matrix. It does not require a priori knowledge of upper bounds of the parameters and of the perturbations and it is expected to improve robustness of current adaptive control algorithms. Two simulation experiments illustrate the efficiency of the proposed algorithm.