Aiming at the shortcomings that the Empirical Mode Decomposition (EMD) part in the combination forecasting method is not insufficient in processing non-linear and non-stationary signal, a short-term wind power prediction method based on hybrid Improved Ensemble Empirical Mode Decomposition (IEEMD) and Least Squares-Support Vector Machine (LS-SVM) model was proposed. Firstly, through the study of the added-noise principle of noise assisted decomposition method, the additive noises applied in the form of positive and negative pairs were deduced to effectively eliminate the residual noise within the components, and the two additive noises parameters of the amplitude of additive white-noise and the number of ensemble trials were determined to fixed as 0.014 times standard deviation of the original signal and two ensemble trials ,respectively .Furthermore, the original data was decomposed into a series of Intrinsic Mode Functions (IMFs) by IEEMD method, which screened and restructured into three frequency range components with high frequency, intermediate frequency and low frequency by the run-lengths test. And then, those components with different frequency bands were established for the LS-SVM multi-step prediction models. Finally the prediction values were adaptively superposed to obtain the predicted result. Through the simulation experiments and the measured wind power experiments verify that the proposed method has certain advantages in prediction accuracy, which also provides a novel idea for the short-term prediction method.