Abstract: To address the threats posed by the strong fluctuations and intermittency of wind power output to the safe and stable operation of the power system and the increased difficulty of grid regulation, a hybrid energy storage capacity allocation strategy based on CEEMDAN-NGO-VMD is proposed to stabilize wind power fluctuation. Firstly, based on principal component analysis dimensionality reduction and fuzzy C-means clustering, typical days of the annual output data of the wind farm are extracted. Secondly, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is adopted to decompose the original wind power signal into the direct grid-connected component that meets the fluctuation limit and the component that the hybrid energy storage system (HESS) can bear, achieving the overall wind power stabilization goal. Then, the internal power of the HESS is divided by using the Northern Eagle optimized variational mode decomposition algorithm. On this basis, the high and low frequency boundary points are determined according to the energy entropy difference of the adjacent eigenmode functions obtained by decomposition, and the power borne by the lithium battery and the supercapacitor respectively is obtained. Finally, an HESS capacity optimization allocation model for stabilizing wind power output fluctuations is established, and the model is solved based on the extracted typical daily data. The results of the calculation example show that the CEEMDAN method adopted in the first layer decomposition effectively mitigated the wind power fluctuations. The maximum wave momentum of the grid-connected power after stabilization within 10 minutes was only 42.11% of the national standard. Meanwhile, the double-layer decomposition strategy, through the refined allocation of the output of the energy storage system, enables the HESS capacity obtained through configuration to have better economic efficiency.