Abstract: To solve the problem that wind power is difficult to predict accurately due to intermittent and nonlinear fluctuations, a wind power prediction model based on“sequence decomposition-network prediction-sequence reconstruction”is proposed. First, according to the output characteristic curves of various wind turbines in wind farm clusters, the typical output curves are obtained through Iterative Self Organizing Data Analysis Techniques Algorithm (ISODATA) clustering algorithm. Then, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is used to decompose the original wind power series data obtained by clustering to reduce the prediction error caused by data fluctuation. Third, the Bi-directional Long Short-Term Memory (BiLSTM) prediction model of each modal component is established, the Improved Sparrow Search Algorithm (ISSA) is used to optimize the network parameters, and the final prediction result of wind power is obtained by superimposing the prediction results of each modal component. The numerical results show that the prediction accuracy of the proposed model is higher than that of other models, and it has better generalization ability.

    Key words: wind power prediction; complete ensemble empirical mode decomposition with adaptive noise; bi-directional long short-term memory; improved sparrow search algorithm