Abstract: A short-term wind speed forecast model combining the weather research and forecast（WRF）and attention mechanism is proposed. Firstly,the WRF model is applied to simulate multiple data,including wind speed,wind direction,temperature,and humidity,which are regarded as input variables of the subsequent algorithm. Secondly,the errors of the wind speeds resulting from WRF and the other related variables are decomposed into sub-modal components with various frequencies by the variational mode decomposition（VMD）algorithm to reduce the complexity and non-stationary feature of the raw data. Thirdly,as for Bi-directional Gate Recurrent Unit（BIGRU）with attention mechanism（AM）,the adaptive grid search algorithm is adopted for the parameter optimization of the model structure. Finally,based on the proposed model,the errors are forecasted to correct the simulated wind speed of WRF.Through the analysis of examples,the accuracy of the proposed model is better than that of the comparative models in single-step and multi-step forecast,which proves the superiority of the model.