The accuracy of wind power prediction is of great significance for optimizing power system scheduling and improving power output stability. In order to solve the established difficulties in wind prediction, a prediction method combining multiple models is proposed with the aim of realizing a high-precision wind prediction strategy up to 1 hour in advance. Firstly, the raw wind power data is preprocessed. Secondly, the comprehensive feature se-lection (CFS) algorithm is used to calculate the sum of the importance of each feature variable, and the input set of asynchronous eldest child models corresponding to the target feature is selected. Then, the adaptive noise complete set empirical mode decomposition algorithm (CEEMDAN) is used to decompose the wind power sequence, and each component after decomposition is modeled separately. The comparative analysis of a series of experimental methods shows that the method proposed in this study demonstrates excellent prediction accuracy and stable performance in the field of multi-step wind energy prediction.