For the end effect of ensemble empirical mode decomposition（ EEMD）,considering the disadvantages that the extreme extension only uses the value near the endpoints and the support vector regression（ SVR） extension based on data points takes much time,a new SVR extension method based on extreme points is proposed by the combination of extreme extension and SVR extension. Firstly,extreme points near the endpoints are used as the SVR training samples to extend extreme value.Secondly,the average difference between the original signal time scales of the adjacent extreme points is used to control the shape of these extended extreme points. Finally,these points are interpolated to the required data points by Hermite. The method refers to the variation of signal endpoints,meanwhile,considers the information inside extreme points of the entire data sequence.Simulation and experimental short signal results show that SVR extension method based on extreme points not only can improve the accuracy and reliability of the EEMD decomposition,depress end effect effectively and solve the decomposition distortion caused by end effect,but also sharply reduce the number of SVR extension data points,distinctly shorten the time of SVR extension and increase the practicability of the method.