Convergence Improvement for Multi-Individual Optimization based Identification Using Output Over-Sampling

Abstract

System identification suffers from severe numerical problems when the experimental data do not contain sufficient independent information on system dynamics, or they are contaminated by complicated noises. As a result, numerical optimization used in identification cannot guarantee the global optimal solutions, thus most of the existing identification algorithms have poor global convergence performance unless strong external test signals are available. In this paper, a new identification algorithm based on multi-individual optimization is investigated to improve its convergence performance under severe numerical conditions. It is illustrated that some distinctive information can be extracted from the experimental data collected by output over-sampling scheme, and can complement information criterion for numerical optimization. Furthermore, the multi-individual scheme is utilized to decrease the influence of initial conditions and local minima. The numerical simulation examples illustrate that the convergence performance has been improved in the proposed algorithm.