小波网络在带噪声的混沌时间序列预测中的研究

在采用网络模型对带有噪声的混沌时间序列进行建模的过程中,噪声会影响模型的泛化能力。针对上述问题,本文提出了基于小波去噪的小波网络预测框架。在预处理阶段使用小波阈值方法抑制噪声,运用相空间重构理论确定嵌入维数和延迟时间,进而确定改进的小波网络模型的结构,结合BP算法和遗传算法对模型的参数进行学习。最后,在带噪声的Mackey-Glass混沌序列预测实验中验证了该框架的有效性。

Prediction Research of the Chaotic Time Series with Noise Based on Wavelet Neural Networks

In the process of modeling chaotic time series,because noise will weaken the generalization ability of the neural network model,a forecast framework is proposed to solve this problem.Wavelet denoise is introduced in the preprocessing stage to depress the influence of noise,then the embedded dimension and delay time are determined by the phase space reconstruction theory to choose the best structure of hybrid wavelet neural networks.Finally the hybrid wavelet neural network model is trained by both the genetic algorithm and the back-propagation one.Experiments on the Mackey-Glass chaotic time series validate the framework.