Optimal embedding parameters: a modelling paradigm (Q1888128)

The authors study the problem of an optimal embedding for the reconstruction of a dynamical system from a (noisy) time series from a modelling perspective. Their main point is that for modelling the evolution operator only the product of embedding dimension and embedding lag, called the embedding window, are relevant and not the individual values of these parameters. They also introduce an information theoretic criterion for the optimal selection of the embedding window. They demonstrate the performance of their approach first at data obtained by a numerical integration of the Rössler and the Lorenz system, respectively, and from the Ikeda map. Finally, the application to three experimental time series is discussed: the annual sunspots numbers, human electrocardiogram recordings of ventricular fibrillation and laser intensity data.