一种基于支持向量回归机的经验模态分解方法

经典经验模态分解中采用基于曲线参数插值拟合局部均值曲线,该类方法对参加拟合的极值点很敏感,若出现干扰的异常极值点或得不到真实的极值点,将导致分解结果失真或产生模态混叠.提出一种基于支持向量回归机拟合局部均值曲线的经验模态分解方法,即利用序列的极值点训练支持向量回归机拟合局部均值曲线代替传统的曲线参数插值.实验表明,与经典算法相比,该方法具有更好的频率分辨率,对采样频率不敏感且能克服微弱高频间断信号的干扰,有效解决Hilbert-Huang变换中存在的模态混叠问题.

Empirical mode decomposition based on support vector regression machines

Empirical mode decomposition(EMD),which fits the curve of local mean values based on parametric interpolation,is sensitive to the extremal points involved in fitting.When abnormal or false extremal points are encountered,the result will suffer from distortion or modal mixture.An effective EMD method based on support vector regression machines is proposed to fit the curve of local mean values,which uses the extremal points of a time sequence to train the support vector regression machine to fit the curve of local mean values in place of the curve fitted by traditional parametric interpolation.It is found that the proposed algorithm has higher frequency resolution,lower sensitivity to sampling frequency and no interference from weak high-frequency discontinuous signals,and overcomes the problems of modal mixture in Hilbert-Huang transform.