The accuracy of wind speed forecasting is related to the wind power scheduling. When large-scale wind power connected grid, it also affects the stability of the grid. This paper applies time series model and Back Propagation (BP) neural network model to predict wind speed. Finally, a combination model of time series and BP neural network is proposed. In the combination model, the inputs of BP neural network are made up of historical data and residual errors calculated by time series model. The model can be more accurately in the short-time wind speed forecasting. And then shows an actual example.