The invention discloses an ultra-short-term wind power combination prediction method based on a support vector machine, and the method comprises the steps: firstly carrying out the linear interpolation replacement of to-be-processed wind power historical data according to the data of an adjacent time period, and carrying out the normalization of the preprocessed data; secondly, decomposing the processed wind power data into an eigenfunction sequence and a residual error sequence by using empirical mode decomposition; secondly, establishing a quantum particle swarm-support vector machine model for the eigenfunction sequence and the residual sequence obtained by decomposition, and performing training optimization to obtain a predicted value of each sequence; and finally, superposing the prediction values of the sequences to obtain a final wind power prediction value, and carrying out error evaluation analysis. Compared with a support vector machine direct prediction result or a result without data feature decomposition, the prediction result of the method is improved, and meanwhile the situation that local errors are too large does not occur. Compared with an existing wind power prediction scheme, the method is higher in robustness, higher in calculation speed, less in data requirement and better in prediction effectThe invention discloses an ultra-short-term wind power combination prediction method based on a support vector machine, and the method comprises the steps: carrying out the linear interpolation replacement of to-be-processed wind power historical data according to the data of an adjacent time period, and carrying out the normalization of the preprocessed data; secondly, decomposing the processed wind power data into a cost characteristic function sequence and a residual sequence by utilizing empirical mode decomposition; secondly, establishing a quantum particle swarm-residual sequence for the intrinsic function sequence and the residual sequence obtained by decomposition; carrying out training optimization on the support vector machine model to obtain a predicted value of each sequence; and finally, superposing the predicted values of the sequences to obtain a final wind power predicted value, and carrying out error evaluation analysis. Compared with a result of direct prediction of a support vector machine or no data feature decomposition, the prediction result of the method is improved, and meanwhile, the situation of overlarge local error does not occur. Compared with an existing wind power prediction scheme, the method is higher in robustness, higher in calculation speed, less in data demand and better in prediction effect.