Echtzeit-Identifikation nichtlinearer instationärer Systeme mit RBF-Netzwerken

The paper focuses on the application of artificial neural networks (ANN) for modelling of nonlinear dynamic, stationary and non-stationary systems. It is shown that radial basic function nets (RBF) in connection with the newly developed fast multistep algorithm give the best identification results. The structure of the net is automatically extended during the learning process, until the error is below a defined limit for all training data samples. The characteristics of the new identification method are illustrated and compared with other well-known methods with an example of a nonlinear difference equation of first order with a time-variant factor