An Embedding Selection Algorithm for Chaotic Dynamical Systems

The selection of an embedding scheme is an important step in the modeling and prediction of chaotic dynamical systems. Theoretical work in embedding selection abounds in the literature. However in neural network research, mostly compute intensive methods for embedding selection exist. In this paper, we propose a novel embedding selection scheme based on cluster analysis. A neural network implementing this method is described and demonstrated on the MackeyGlass chaotic time series. The result of the method agrees with the embedding schemes used by researchers in neural networks. In addition, other new embedding schemes have been Found and they also enable this chaotic time series to be predicted accurately.