混沌时间序列的自适应正交小波神经网络预测

Prediction of the Chaotic Time Series Using Adaptive Orthogonal Wavelet Neural Network

针对非线性混沌时间序列预测问题,提出一种基于正交小波神经网络的自适应预测算法。根据来自非线性序列模型的期望输入输出数据,利用小波框架理论建立初始的小波神经网络。采用正交化逐步选择方法对于初始小波神经网络进行结构优化,从而建立最精简的网络模型。同时引入在线学习算法在线修改网络权值和小波神经元的参数,从而提高模型的自适应能力和泛化能力。通过对时滞Mackey-G lass超时间序列和时变Lorenz混沌序列的预测,证明了算法的有效性。

With an aim at the prediction problem of nonlinear chaotic time series, a novel adaptive predictive algorithm based on orthogonal wavelet neural network is suggested. Based on the desired inputoutput data coming from the nonlinear series models, the initial wavelet neural network is established using the wavelet framework theory. The orthogonal stepwise selection method is adopted to optimize the initial wavelet neural network, whereby the simplified network model is established. And at the same time, the on-line learning algorithm, on-line revised network weight values and the parameters of wavelet neural elements are introduced to improve the adaptability and pan-capability. The algorithm effectiveness is proved via the prediction of the time delay Mackey-Glass overtime series and time variation Lorenz chaotic series.