A combination ultra-short-term prediction method of wind power based on analytical mode decomposition (AMD) and improved cuckoo search algorithms optimized support vector machines (ICSA-SVM) is proposed to treat with the randomness, volatility and uncertainty of wind power. Firstly, the wind power is decomposed into components with different frequencies by using AMD to reduce the influence between different frequencies. Then, according to the characteristic of each sequence, different penalty parameters and kernel function parameters are found by using ICSA to improve the forecasting accuracy of single model. Finally, the prediction results are superimposed for error analysis. Simulation results show that the proposed strategy can track the change of wind power better and improve the forecasting accuracy of wind power effectively.