基于稀疏Volterra滤波器混沌时间序列自适应预测

针对混沌时间序列预测问题,提出了一种稀疏Volterra滤波器,该滤波器采用归一化最小均方自适应算法进行多次训练,每次训练只保留滤波系数矢量和输入信号矢量的有效分量进行下一次训练,从而将Volter-ra滤波器的有效滤波系数个数减至最少,降低了预测模型的复杂性。四种混沌时间序列的预测实验表明:该滤波器可同时实现对混沌流和混沌映射的建模与预测,可有效地减少滤波器的滤波系数个数,能在不损失预测精度的前提下,降低预测模型复杂性。

Nonlinear adaptive prediction of chaotic time series based on sparse Volterra filter

A sparse Volterra filter with application to nonlinear adaptive prediction is proposed to investigate prediction performance of four kinds of chaotic times series.The filter coefficients are updated by using normalized least mean square(NLMS) adaptive algorithm.The number of valid filter coefficients is reduced gradually during the updating process,and the complexity of the prediction model is simplified.Experimental results show that the sparse Volterra filter can be successfully used to predict both chatoic flow and chaotic map,can reduce the number of filter coefficients effectively and has the same prediction performance as Volterra filter,simplify the complicacy of the predictor.