A robust off-line output error method for system identification

The model reference technique and Huber's minimax principle have been successfully used to develop an offline output error method for robust identification of systems. This method is named the robust iterative output error method with modified residuals. A convergence analysis of the proposed method has been included as well as some simulation results. In the presence of a small number of large errors (called outliers) in the input-output data, the presented method has demonstrated its distinctive advantages over not only the nonrobust methods but also previously developed robust methods. The main advantages are a fast convergence speed and satisfactory robustness. It is concluded that the method developed here is much superior to the other methods and therefore can be widely used in many real-time applications.<
>