To solve the problem of low wind power prediction accuracy caused by uncertainty and volatility of wind speed, a combined ICEEMDAN-VMD-IGWO-BILSTM model based on ICEEMDAN-VMD, improved Grey Wolf algorithm (IGWO) and BiLSTM was proposed. Firstly, the number of modal components after combinatorial modal decomposition is limited to avoid insufficient decomposition. Secondly, lens imaging reverse learning, nonlinear convergence factor and fusion with Cauchy variation algorithm are introduced to improve the Grey Wolf algorithm to optimize the bidirectional long short-term memory neural network. Then, IGWO-BiLSTM prediction model is established for each decomposed sub-sequence, and finally the predicted value of each sub-sequence is superimposed to obtain the final predicted value. The model is applied to actual wind power prediction. Experimental results show that MAE, MSE and RMSE of ICEEMDAN-VMD-IGWO-BiLSTM combined model are 4.9189MW, 32.3683MW and 5.6893MW, respectively. Compared with CNN-LSTM, CNN-BiLSTM neural network model and other combined models, the prediction accuracy has been significantly improved, which can effectively solve the problem of low prediction accuracy of wind power.