Konferenční příspěvek

Král L., Punčochář I., Duník J.

Proceedings of the 10th International PhD Workshop Young Generation Viewpoint

Hluboka nad Vltavou, Czech Republic

2009

Ing. Ivo Punčochář, Ph.D. (FAV Západočeská univerzita, katedra kybernetiky)

clanek_PhDW09_LK.pdf [file]

prezentace_PhDW09_LK.pdf [file]

A functional adaptive control for nonlinear stochastic Multi-Input Multi-Output (MIMO) systems is presented. A nonlinear system is modelled by a MultiLayer Perceptron (MLP) neural network. Parameters of the model are estimated by the extended Kalman filter (EKF). One of the key problems connected with neural network is selection of its structure. In order to avoid this problem, an on-line algorithm for dynamic structure optimization of the MLP network is proposed. Controller design is based on bicriterial dual approach that uses two separate criteria to introduce opposing aspects between estimation and control; caution and probing. The proposed approach is compared with two adaptive non-dual controllers. The quality of the proposed functional adaptive controller is illustrated in a numerical example.