Data based models: an automatic method for model structure determination

摘要

A technique for automatically determining the structure of linear discrete-time models from real data is briefly illustrated. For pole-placement control system design, as shown in an example, it is essential that near, or exact, pole-zero cancellations should not occur within the model. It has been found that the Young's information criterion (YIC) statistic has consistently identified model structures that are free from such cancellations, and that subsequently result in good control system performance. Naturally, some control system design algorithms, for example predictive controllers, are insensitive to overparameterised models; however the YIC statistic is still useful, as it normally indicates the simplest model structure, which should result in minimum complexity controllers being designed.