The invention relates to an LSTM-Attention network-based wind power grid control method. The method comprises the following steps that: S1, data preprocessing is performed on different wind speed data; S2, a total network model is established; S3, the total network model is trained with historical data; S4, a wind electricity power result is obtained through the trained network model; and S5, thewind electricity power result is verified, different control operation is performed on a wind power grid according to measures corresponding the wind electricity power result. A prediction model is divided into two parts, namely, the LSTM network of an attention mechanism and an SE_CNN network. Firstly, standardization processing is performed on original NWP data, and the processed data are adopted as the input of the prediction model; secondly, the LSTM network of the attention mechanism and the SE_CNN network are utilized to extract the global features and local features of time sequence NWPdata respectively, then perform feature fusion, and a classifier is utilized to predict a wind speed; and finally, wind electricity power is obtained on the basis of relationships between historicalwind speeds and wind electricity power. Compared with the prior art, the method of the invention has the advantages of high accuracy, high speed and the like.