Avtor/Urednik		Stankovski, Vlado; Debeljak, Žiga; Rape, Rok; Fefer, Dušan
Naslov		Application of two neural network architectures to time series forecasting
Tip		članek
Vir		In: Solina F, Zajc B, editors. Zbornik 2. elektrotehniške in računalniške konference ERK'93. Zvezek A. Elektronika, telekomunikacije, avtomatika, močnostna elektrotehnika, merilna tehnika - ISEMEC 93; 1993 sep 27-29; Portorož. Ljubljana: Slovenska sekcija IEEE,
Leto izdaje		1993
Obseg		str. 584-7
Jezik		eng
Abstrakt		This paper shows an application of two neural networks to time series forecasting. At the beginning analysis of the sampled time series is made with the Grassberger-Procaccia and Kennel-Isabelle method in order to distinguish noise from possible deterministic, time series. This is important because of the presence of noise in fhe observed series. According to the results of the analysis, appropriate time series with deterministic features is selected. A specially designed back-propagation neural network and a neural network proposed by l. Grabec were trained with the selected series. The neural network, proposed by l. Grabec proved to give better prediction results in a short time interval while the BP neural net performed better a few steps ahead in time. The results are encouraging. i I
Deskriptorji		NEURAL NETWORKS (COMPUTER) FORECASTING