Optimal embedding parameters: A modelling paradigm

摘要: The reconstruction of a dynamical system from time series requires the selection two parameters: embedding dimension de and lag τ. Many competing criteria to select these parameters exist, all are heuristic. Within context modelling evolution operator underlying system, we show that one only need be concerned with product deτ. We introduce an information theoretic criterion for optimal window dw = For infinitely long series, this method is equivalent selecting minimises nonlinear model prediction error. short noisy find results new algorithm data-dependent superior estimation standard techniques.