Abstract: A novel multi-step forecasting model of ultra-short term wind power based on the combination of the enhanced colliding bodies optimization (ECBO), the variational mode decomposition (VMD) and the wavelet kernel extreme learning machine (WKELM) is proposed. The ECBO algorithm is used to optimize the key parameters of the VMD, herein, a new fitness function is designed based on the idea of the weighted-permutation entropy (WPE), which not only improves the self adaptability of the VMD method, but also realizes the quantitative discrimination for the regularity of each decomposition component. First, the original wind power time series is decomposed by the ECBO-VMD adaptively, then the forecasting model based on the WKELM is established for each decomposition component, and the final forecasting results are obtained by reconstructing the forecasting value of each component. The experimental results confirm that the proposed method can significantly improve the multi-step forecasting accuracy compared with the existing single or combined forecasting methods, and the error distribution of the forecasting values can be controlled in a narrower expected forecasting range.