Study on the Evaluation of Phase Space Reconstruction Based on Radial Basis Function Network Prediction

Abstract

The radial basis function neural network model was applied to characterize the nonlinear dynamical behavior of the chaotic attractor neighborhood evolution, and the key problem of how to ensure the reconstructed dynamical system is identified with original one topologically by the calculated reconstruction parameters was solved effectively. The proposed method characterizes the nonlinear dynamics more accurately than the traditional linear model prediction. Simulation results show that the original phase space can be reconstructed by the suitable embedding dimension and delay time, and the short-term prediction of chaotic time series is realized by the radial basis function network. The method was reasonable for proving the availability of phase space reconstruction.