Author/Editor		Stankovski, Vlado; Fefer, Dušan
Title		Prediction and system identification supported by measures of chaos
Type		članek
Source		In: Keating JG, editor. Neural computing: research and applications 3. Proceedings of the 5th Irish neural networks conference; 1995 Sep 11-13; Maynooth. St. Patrick's College: Maynooth,
Publication year		1995
Volume		str. 70-5
Language		eng
Abstract		We observe an AC/DC transfer standard by measurements of relevant variables. The obtained data is in the form of time series. Our main objective is to see whether the maesured time series have some determinisstic properties, which we call system identificatioon, and if that is the case to see how good it is possible to predict the future behaviour of the system. We show that it is difficult to identify the observed real system by using the method proposed by Grassberger and Procaccia. Then we construct back-propagation neural networks of the future values of the time series for several steps ahead in time. we show that by computing the correlation coeffiicents between the actual and the predicted values in the same time series and by drawing them on a graph it is possible to identify the observed system. Neural networks may be also capable of demonstrating a great degree of accuracy in predicting the behaviour of such systems.
Descriptors		NONLINEAR DYNAMICS NEURAL NETWORKS (COMPUTER) FORECASTING