In order to improve the accuracy of short-term wind direction forecasting, a hybrid model, named EEMD-CNN-GRU, is proposed based on ensemble empirical mode decomposition (EEMD), convolutional neural network (CNN) and gated recurrent unit (GRU). According to the characteristics of randomness and unsteadiness of wind direction series, EEMD is used to decompose the data into multiple components. Then, CNN’s local connection and weight sharing are harnessed to extract the potential features in the components. Finally, GRU is adopted to further construct features of the potential features extracted by CNN, and the predicted values of each component are superposed to obtain the ultimate prediction results. The experimental results show that the proposed prediction method achieves good performance compared with other models such as BP neural network and long short-term memory (LSTM).