Abstract: To obtain reliable and accurate wind speed data, a data cleaning method for measured wind speed of wind turbines was proposed in this study. The method incorporates spatiotemporal correlation by utilizing a graph convolutional neural network（GCN） to extract spatial correlation information and a bidirectional long short-term memory neural network（Bi-LSTM） to extract temporal correlation information. A GCN-LSTM model was established to reconstruct the wind speed of each wind turbine, so as to realize identification and removal of abnormal wind speed. The study also analyzes the spatiotemporal characteristics of wind speed and their impact on the accuracy of the proposed model. Two important modeling parameters are identified: the optimal time scale and the number of wind turbines. The proposed method was validated by using data from four wind farms with different terrains in China. The results show that incorporating spatiotemporal correlation can effectively improve accuracy of data cleaning. Moreover, the higher the spatiotemporal correlation of wind speed, the smaller the cleaning error. The proposed model has robustness in cleaning wind speed data under various terrain types.