Abstract: This paper proposes a new short-term wind power prediction model based on convolution long-short-term memory neural networks（CNN-LSTM）,which uses wind power historical data and numerical weather prediction（NWP）data including wind speed and wind direction as the input information. Firstly,the original multi-dimensional input variables are preprocessed by principal component analysis（PCA）,and the principal components of the input variables are selected as the input of the neural networks. Then the processed data is extracted by the convolutional networks,and is next fitted through the long-short-term memory networks. The DropConnect technique is introduced in the training process of the neural networks to reduce the over-fitting phenomenon in the model.Finally,accurate wind power prediction is achieved. This model is validated by predicting a real wind farm output in Northwest China.The results show that the proposed method can effectively predict wind power and has higher prediction accuracy than that by BP neural networks and support vector machine（SVM）.