Abstract: Aiming at the problems of poor prediction accuracy and poor adaptability of different combination algorithms in the current short-term forecasting of wind speed series, this paper proposes a combination forecasting model based on wavelet transform, which reduces the volatility and disorder of wind speed series through wavelet transform. The shuffled frog leaping algorithm（SFLA） is used to optimize the initial weight and threshold of the back propagation（BP） neural network, and the difference evolution（DE） algorithm is used in the SFLA’s subpopulation individual optimization strategy which improves the speed and accuracy of individual convergence. The high and low frequency components decomposed by the wavelet transform are respectively used for wind speed prediction and reconstruction through the combined model algorithm. Compared with the 60 min prediction results, the mean absolute percentage errors of 10 min and 30 min were increased by 33.59% and 12.21% respectively, and the root mean square errors were increased by 28.77% and 8.22% respectively. The average prediction errors of the three are 0.037,-0.014, 0.011 m/s, horizontally compared with the SFLA-BP neural network algorithm and the BP neural network algorithm, the results show that the combination forecasting model of this paper predicts the best performance indicators.