《On-line Identification of Nonlinear Time-Variant Systems - Thomas F. Junge》.pdf

Proceedings of the American Control Conference Albuquerque, New Mexico June 1997 0-7a03-3a32-41971$10.00 o I 997AACC On-line Identification of Nonlinear Time-Variant Systems Using Structurally Adaptive Radial Basis Function Networks Thomas F. Junge and Heinz Unbehauen Control Ehgineering Laboratory (ESR) Faculty of Electrical Engineering, Ruhr-University Bochum D-44780 Bochum, Germany Phone: ++49 234 700 4072 Telefax: ++49 234 709 4101 E-mail: jungeOesr.ruhr-uni-bochum.de Abstract storage capacity makes this approach impossible for real-time applicati- ons. Thispaper presents a new algorithm to train Direct Linear Feedthrough Radial Basis Function (DLF-RBF) networks, especiallydesignedf i r on- The aim in this paper is to develop a training algorithmthat incorporates line identification of time-variant nonlinear dynamical systems. The the idea of on-line structural adaptation, and adapting the RBF center algorithm basically explores the network’s input space and the model positions and the widths using an error sensitive clustering algorithm error to determine automatically the number of RBF neurons, and to which considers the shape of the local approximation (adaptive error adapt their centerpositions (adaptive error dependent clusterind. The dependent clustering).The output layer weights are adapted using a RL