Predictions models with neural nets

Acta Universitatis Agriculturae et Silviculturae Mendelianae Brunensis

Vladimír Konečný¹ 

[1] Ústav informatiky, Mendelova zemědělská a lesnická univerzita v Brně, Zemědělská 1, 613 00 Brno, Česká republika;

关键词: trend;    neural net;    time series;    model;    learning of neural net;    error function;   

DOI  :  10.11118/actaun200856030109

来源: DOAJ

【 摘 要 】

The contribution is oriented to basic problem trends solution of economic pointers, using neural networks. Problems include choice of the suitable model and consequently configuration of neural nets, choice computational function of neurons and the way prediction learning. The contribution contains two basic models that use structure of multilayer neural nets and way of determination their configuration. It is postulate a simple rule for teaching period of neural net, to get most credible prediction.Experiments are executed with really data evolution of exchange rate Kč/Euro. The main reason of choice this time series is their availability for sufficient long period. In carry out of experiments the both given basic kind of prediction models with most frequent use functions of neurons are verified. Achieve prediction results are presented as in numerical and so in graphical forms.

【 授权许可】

Unknown