In order to improve the accuracy of wind power prediction, a short-term wind power combination prediction model based on CEEMDAN decomposition based on SMA optimization LSSVM was established. First, the original wind power sequence was decomposed and reconstructed by using full set empirical mode decomposition. Then, according to the new sequence, the corresponding prediction model was established. In order to optimize the pa-rameters of the least square vector support machine model, the slime mold algorithm was proposed to optimize and improve the model performance by adjusting the parameters of LSSVM. Finally, a variety of comparison models were constructed for comparison and analysis. The results showes that CEEMDAN-SMA-LSSVM have the highest prediction accuracy and the prediction results are closer to the real value. The research can be used to predict the short-term use of wind power in wind farms.