Abstract: Aiming at the problem of convolution operation is not suitable for extracting spatial correlations of multiple wind turbines with irregular layout in wind speed prediction tasks for multiple wind turbines,a multi-step wind speed prediction algorithm of multiple wind turbines based on the spatial-temporal attention-Seq2Seq model is proposed. Firstly,the spatial attention mechanism is used to strengthen the spatial correlation of wind speed sequences,and the conventional spatial attention mechanism is improved. Then,the encoder in the Seq2Seq model is used to encode the data. Finally,the decoder combined with the temporal attention mechanism is used to calculate the multi-step prediction results of multiple wind turbines. A wind speed dataset of a wind farm in Hebei is used as a case,the results show that the average absolute error of the proposed algorithm is reduced by about 4.3%-15.0% compared with other comparison algorithms,and the accuracy is significantly improved.