Initialization of output error identification algorithms
Elie Tohme · OpenGrey (Institut de l'Information Scientifique et Technique) · 2008
This thesis deals with the initialization of output error identification algorithms in order to achieve the global convergence. Two approaches are proposed. The first approach is based on the reinitialized partial moments. This equation error type approach is compared with several methods using stochastic simulations in both discrete and continuous domains. The obtained results show the efficiency of the reinitialized partial moments to initialize output error algorithms. In the second approach, the proposed algorithms, named pseudo-output error algorithms, are built from the standard output error methods by trying to satisfy a particular positive realness condition that assures the global convergence. Despite the bias of the estimated parameters, these approaches provide a good initialization to the output error algorithms.