Abstract: To improve the accuracy of wind power forecasting, an ultra-short-term wind power combined prediction based on the complementary ensemble empirical mode decomposition (CEEMD) the satin bower bird optimization algorithm (SBO) and the optimized least squares support vector regression (LSSVR) is proposed. With the random volatility of wind power sequences, the CEEMD is used to decompose the wind power data, and the decomposed series of components with different time characteristic scales are used as the training inputs for the LSSVR model. Then SBO is introduced to optimize the regularization parameter and the width of the kernel function of the LSSVR, and the wind power prediction model is established for each component. The final predicted values can be obtained by superimposing the prediction value of each component. The combined prediction based on the CEEMD-SBO-LSSVR not only effectively reduces the complexity of prediction, but also ensures the original wind power sequence's a small reconstruction error after the modal decomposition. The simulation results illustrate that this method has higher prediction accuracy for ultra-short-term wind power than other predictions.