A general criterion is proposed for robust identification of both linear and bilinear systems. Following Huber’s minimax principle, the ordinary iterative Gauss-Newton approach is applied, with modified residuals, to minimize the suggested robust cost function. The proposed method, named the robust iterative least squares method with modified residuals (RILSMMR), can provide simultaneously robust estimates of the system parameters as well as the residual variance. Therefore it is superior to the earlier robust methods. A proof of convergence of the RILSMMR is given. Results of simulation with both the RILSMMR and nonrobust identification methods are included. These confirm that RILSMMR has certain advantages over both conventional nonrobust identification methods, as well as earlier robust methods.

This content is only available via PDF.
You do not currently have access to this content.