Abstract: To address the demand for smoothing grid-connected wind power caused by the inherent volatility and randomness of wind power, a lithium battery-flywheel hybrid energy storage system was configured on the wind-farm side, while a wind power fluctuation mitigation control strategy was proposed with considering ultra-short-term wind power prediction, model predictive control (MPC)-based rolling optimization, and hybrid energy storage power allocation. First, an attention mechanism (AM) was incorporated into long short-term memory (LSTM) network for ultra-short-term wind power prediction, and the prediction results were fed as disturbance inputs into MPC model for correction, thereby improving the model prediction accuracy. Second, the MPC method was employed to obtain the grid-connected wind power and the target smoothing power of the energy storage system, so that the wind power output has been smoothed while the stability of the state of charge (SOC) of the energy storage system has been maintained. Then, based on the different characteristics of hybrid energy storage components, the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) method was applied to reasonably allocate the real-time power between the lithium battery and the flywheel. Finally, simulation analysis was carried out using actual wind farm power data. Results show that the battery SOC operating state can be significantly improved by the proposed control strategy, the wind power smoothing requirements of the energy storage system are met while the energy storage lifetime is taken into account, sufficient charge/discharge margins of the energy storage system are ensured to cope with possible future changes in the system, and the frequency of abrupt SOC variations is reduced.