Abstract: The traditional ultra-short-term wind speed prediction often utilizes the wind speed signals at a single location within the wind farm, ignoring the correlation of the wind speed among wind turbines. It is difficult to consider the spatial distribution characteristics of the wind speed under the influence of terrain and wake, thus affecting the improvement of ultra-short-term wind speed prediction accuracy. Therefore, a multi-location ultra-short-term wind speed prediction method based on the combination of Recurrent and Convolutional Neural Networks is proposed in this paper, which can predict wind speed spatial distribution under consideration of the spatial and temporal correlation of wind speed. The proposed method combines the convolutional neural network and the long-short term memory network. The convolutional neural network is employed to obtain long-term wind speed spatial distribution characteristics, while the long-short term memory network is used to get the short-term time series features. With these two networks combined, this method can finally produce the multi-location ultra-short-term wind speed prediction results simultaneously. With the data of a wind farm in Shandong province as an example, it is proved that the accuracy of the proposed model is improved compared with the traditional methods. The test set is divided according to the quarters of a year. In the fourth quarter where the prediction results of all models are the best, for the average error level of multiple points, the MAE and RMSE of the proposed model are 0.367m/s and 0.506m/s, which are respectively reduced by 15.0% and 15.2% than the persistence method model, and reduced by 31.5% and 43.1% than the SVM model.