Chaotic time series prediction and its application based on wavelet neural network

A method for chaotic time series prediction based on wavelet neural network is discussed by analyzing the theory of phase space reconstruction. G-P algorithm and Takens theory are applied to calculate the minimum embedding dimensions which are required by the phase space reconstruction of chaotic time series and will be used as the number of input nodes. Through the time-frequency analysis, the number of hidden nodes can also be determined on a reliable theoretical basis. Finally, the chaotic time series data from Lorenz simulation signal and rolling bearing vibration signal is used to verify the proposed method. It is found that the proposed wavelet neural network perform well in the chaotic time series prediction, and its results agree well with experimental data with high accuracy over BP network. This paper provides an effective approach with practical engineering significance to the prediction of nonlinear dynamic systems.