Abstract: The empirical mode decomposition is an effective method for non-stationary and nonlinear signals processing, but at the same time it produces endpoint effects which would decrease the decomposition accuracy when the three order spline interpolation is used repeatedly. In order to suppress the endpoint effect in empirical mode decomposition, this paper proposes an approach based on least squares support vector machine and mirror extension. In this new method, by using LS-SVM to the original signal sequence it would produce a finite number of extensional data points at both ends, and then with mirror extension acting on the extensional data a symmetrical continuation signal sequence would be formed. Then a ring signal sequence would be produced after the end extension and mirror extension. The empirical mode decomposition is used to deal with the ring signal sequence. By analyzing the simulation signals and real EEG signals with the new method, the results show that the method can effectively restrain the endpoint effect. And compared with other extension methods, it is better than Support Vector Machine（SVM）, and the Least Squares Support Vector Machine（LS-SVM）.

Key words: empirical mode decomposition, end effect, Least Squares Support Vector Machine（LS-SVM）, mirror extension, Support Vector Machine（SVM）

摘要： 经验模态分解能有效处理非平稳、非线性信号，但在多次采用三次样条插值获取信号上、下包络的过程中容易产生影响分解精度的端点问题。为了抑制经验模态分解中存在的端点效应问题，提出了一种基于最小二乘支持向量机和镜像延拓的端点效应抑制方法。该方法采用最小二乘支持向量机对原始信号序列两端分别向左、右各延拓有限个数据点；用镜像延拓对延拓后的信号序列进行对称延拓处理，将其延拓成一个环形信号序列；对这一环形信号序列进行经验模态分解。通过对仿真信号以及真实脑电信号进行实验分析以及与其他延拓方法的对比，结果表明该方法能够有效抑制端点效应问题，并在抑制端点效应问题方面优于传统的支持向量机和最小二乘支持向量机。

关键词: 经验模态分解, 端点效应, 最小二乘支持向量机, 镜像延拓, 支持向量机