采用支持向量回归抑制噪声的经验模态分解方法

在实际信号分解中，经验模态分解（ EMD）是对噪声敏感的，往往会分离出一些虚假的本征模函数，对信号的分析产生一定影响。为了提高EMD分解的正确率，减少其出现虚假本征模函数的情况，文中提出了一种基于支持向量回归（ SVR）的去噪方法。先对一次EMD分解结果进行SVR逐层滤波并且对信号进行重组，然后利用EMD方法对重组信号进行二次分解。实验表明，二次分解结果已经非常接近于理想的分解结果，不会出现虚假IMF。这种分解方法对噪声不敏感，能有效提高EMD方法对噪声的容忍度。

Empirical Mode Decomposition Method Using Support Vector Regression to Suppress Noise

Empirical Mode Decomposition （ EMD） is sensitive to noise in actual signal decomposition. False intrinsic mode functions tend to exist in decomposition results,leading to negative effects to signal analysis. To improve the accuracy of EMD and reduce the condition of existing the false intrinsic mode function,in this paper,a new de-noising method based on Support Vector Regression （ SVR） . Firstly, decompose the signal with EMD,filtering every IMF by SVR and recombining the regression results. Then decompose the recombined signal with EMD once more time. Experimental results show that the secondary decomposition result is very close to ideal situation and no false IMF is appeared in it. This method is not sensitive to noise,which can effectively improve the tolerance of EMD to noise.