Chaotic time series prediction with feed-forward and recurrent neural nets

• Autorzy: Mańdziuk J. , Mikołajczak R.
• Języki publikacji: EN

Abstrakty

EN

The results of experimental comparison between several neural architectures for short-term chaotic time series prediction problem are presented. Selected feed-forward architectures (Multi-layer Perceptrons) are compared with the most popular recurrent ones (Elman, extended Elman, and Jordan) on the basis prediction accuracy, training time requirements and stability. The application domain is logistic map series - the well known chaotic time series predition benchmark problem. Simulation results suggest that in terms of prediction accuracy feed-forward networks with two hidden layers are superior to other tested architectures. On the other hand feed-forward architectures are, in general, more demanding in terms of training time requirements. Results also indicate that with a careful choice of learning parameters all tested architectures tend to generate stable (repeatable) results.

Słowa kluczowe

EN

chaotic time series; logistic map; neural networks; prediction

PL

odwzorowanie logistyczne; predykcja; sieć neuronowa; szereg czasowy chaotyczny

• Wydawca: Systems Research Institute, Polish Academy of Sciences
• Czasopismo: Control and Cybernetics
• Rocznik: 2002
• Tom: Vol. 31, no 2
• Strony: 383--406
• Opis fizyczny: Bibliogr. 41 poz.,Rys., wykr.,

Twórcy

autor

Mańdziuk J.

autor

Mikołajczak R.

• Faculty of Mathematics and Information Science, Warsaw University of Technology, Plac Politechniki 1, 00-661 Warsaw, Poland,