基于支持向量回归机的HHT边界效应处理

End effects processing in HHT based on support vector regression machines

针对希尔伯特-黄变换中的边界效应,提出了基于支持向量回归机的时间序列预测方法.在支持向量回归机的应用当中,参数的选取对它的泛化性能有很大影响.在讨论了参数对支持向量回归机的泛化性能的影响基础上,提出了通过微粒群优化算法来优化支持向量回归机参数的方法,使得支持向量回归机在应用中能够自适应的选择最优参数,从而获得了更好的泛化性能,提高了在端点处的延拓精度,很好地抑制了端点效应.试验表明,该优化算法能够很好解决支持向量回归机的参数选取问题.通过与神经网络的延拓方法和黄等人的HHTDPS结果对比,基于支持向量回归机的时间序列预测方法可以更好地解决在希尔伯特-黄变换中存在的边界效应,得到的固有模态函数具有较小的失真.

In order to better restrain end effects in the Hilbert Huang transform （HHT）, a time sequence prediction technique is proposed based on support vector regression machines to improve time series prediction. In the application of support vector regression machines （SVRM）, parameter selection has a great influence on generalization performance. So in this paper, the influence of parameters on the generalization of SVRM is discussed, and then a particle swarm optimization （PSO） algorithm is used to optimize parameters. Using this method, SVRM can select optimal parameters self-adaptively, so that higher generaliza tion performance is obtained in applications, prediction accuracy is improved at both ends and the end effects are restrained effectively. In contrast to the neural network methods and HHTDPS proposed by Huang et al. , the end effects can be restrained better and the Intrinsic Mode Functions have less distortion. Experiments show that this method can solve the problem of selecting parameters properly.