Abstract: Aiming at the characteristics of wind speed series such as volatility and randomness,this study develops an ultra-short-term wind speed forecasting method based on an optimal combination model of secondary decomposition and crow search algorithm. The basic idea is to construct a secondary decomposition process based on variational mode decomposition,sample entropy and singular spectrum analysis,which is applied to decompose the original wind speed series into a set of subcomponents. After this,establish a prediction model for each subcomponent. Specifically,for the subcomponents of variational mode decomposition,a deep learning model based on long short-term memory network is applied,and for the subcomponents of residual sequence after secondary decomposition,a combination forecasting model optimized by crow search algorithm is designed. The final prediction is obtained by reconstructing the predictions of all subcomponents. Using real observation datasets of wind speed for model simulation,the developed method decreases MAPE by 19.76%,24.91%,and 26.36% compared with several other models,signifying the effectiveness and stability of the developed method for ultra-short-term wind speed forecasting as well as a good generalization performance on different datasets.