Author/Editor		Stankovski, Vlado; Fefer, Dušan
Title		Prediction and system identification by using neural nets
Type		članek
Source		In: Zajc B, Solina F, editors. Zbornik 3. elektrotehniške in računalniške konference ERK'94. Zvezek A. Vabljena predavanja: eletronika, telekomunikacije, avtomatika, močnostna elektrotehnika, merilna tehnika - ISEMEC 94; 1994 sep 26-28; Portorož. Ljubljana: Slovenska sekcija IEEE,
Publication year		1994
Volume		str. 527-30
Language		eng
Abstract		We consider the problem of prediction of time series exhibiting broadband spectrum which arise from the intrinsic nonlinear dynamics of the system that is observed. The time series are embbeded in a reconstructed phase space which captures the attractor on which the system evolves. The predictor of future points in the phase space is a back-propagation neural network especially designed for this purpose. We trained the neural network with various time series, both sampled from a physical system or computed by integration of differential equations. We examine the prediction results by computing the relative prediction error (Rer) and the correlation coefficient between the real and the predicted time series. We show that by computing the correlation coefficient it is possible to distinquish the deterministic and chaotic from the random time series. This measure can further be used in the process of identification of the observed system.
Descriptors		NEURAL NETWORKS (COMPUTER) FORECASTING ARTIFICIAL INTELLIGENCE