Abstract: In order to improve the prediction accuracy of wind power in new wind farms effectively，the

influencing factors between relevant wind farms were considered. A variational mode decomposition（VMD）

technique was proposed to decompose the wind power preprocessing of a single wind farm into intrinsic mode

function（IMF），and then the same frequency band component such as low-frequency components，high-frequency components and residual components of each wind farm were combined respectively as the input of convolution neural network（CNN）. CNN was used to extract the characteristic information under the same split sub-mode，which was input to the long short-term memory（LSTM）network for prediction，and finally the prediction results were overlaid to obtain the complete prediction results. Compared with a single model，the hyperparameter setting of the combined neural network will affect the prediction accuracy more. A new sparrow search algorithm（SSA）was proposed to save the time of manual parameter adjustment and improve the accuracy and efficiency of hyperparameter setting . The proposed method was used to predict the new benchmark wind farm in a wind power cluster，the result verifies that the VMD-CNN-LSTM optimized by SSA has a higher accuracy in predicting the wind power cluster data，which is higher than the comparison model LSTM，CNN-LSTM and SSA-VMD-LSTM.

      Key words: wind power；variational mode decomposition（VMD）；convolution neural network（CNN）；long

short-term memory（LSTM）network；sparrow search algorithm（SSA）