engleski

Comparison of Several Recursive Identification Methods

In this paper connection between theoretical achievements in the field of recursive identification and practical, applied methods is established. Such an approach should help control engineers in choosing adequate algorithm for an application in industry. Basic definitions introduce notation used. Least squares (LS) and instrumental variable (IV) techniques are described. Several choices for instruments in IV method are given. Recursive identification scheme is presented. Often used Biermann-Thornton factorization algorithm is suited for use with instrumental variable methods. Method for estimation of the DC offset is presented. These methods were ported into C code for testing in simulations and experiments on lab processes. Also, overdetermined instrumental variable method was coded. Simulations included tests with noise introduced in measured signals and tests with parameter change. Overdetermined IV (ORIV) method showed best results in simulation tests. Experiments were performed on physical processes: lab blower process and lab pump plant were recursively identified. Main problem encountered in experiments was DC offset, which had to be removed. The extended LS (ELS) method which includes DC estimation was superior in this case. After DC removal all the other recursive identification methods showed acceptable results.

identification; least squares method; instrumental variable method; factorization

nije evidentirano