To solve the problem of low forecasting accuracy of wind power, a hybrid forecasting model integrating singular spectrum analysis (SSA), convolutional neural network (CNN), bi-directional gated recurrent unit (BiGRU) and Attention mechanism is proposed. In order to suppress the lag problem of the predicted power curve caused by the random fluctuation of wind power, the SSA method is used to decompose the original data into a series of subcomponents with a relatively stable fluctuation characteristics, and the effective reconstruction is realized based on the fuzzy entropy (FE) value of each component. The CNN-BiGRU-Attention model is designed for the forecasting of each reconstructed component. For example, CNN network is used to effectively extract the high-dimensional data features of each reconstructed component, BiGRU network is used to capture the nonlinear dynamic changes of key feature vectors obtained by CNN, and the Attention mechanism is used to highlight and strengthen the effective learning of key features of power data. The final forecasting results are obtained by superimposing the prediction values of each reconstruction component based on the CNN-BiGRU-Attention model. The actual operation data of wind farms in the Hami region of Xinjiang province is used as experimental data, the results show that the proposed method can effectively alleviate the lag phenomenon in wind power forecasting, and the forecasting accuracy is fully ahead of other existed prediction methods.