Long-Term Prediction, Chaos and Artificial Neural Networks. Where is the Meeting Point?

This paper presents the advances of a research using a combination of recurrent and feed-forward neural networks for long term prediction of chaotic time series. It is known that point-to-point, long term prediction for chaotic time series is not possible; however, in this research we are looking for ways to build dynamical systems using artificial neural network, that contain the same characteristics as the unknown systems producing a time series. The created systems should be such that they have chaotic invariants similar to the ones presented in the time series. Being able to create such system, the bounds of the predictions may be defined. We present some general concepts related to chaos, dynamical systems and artificial neural networks and present one model that, if not able to predict yet, is able to autonomously oscillate in bounded limits and in a chaotic way. Some experimental results and future work is also presented. [ABSTRACT FROM AUTHOR]