Abstract: Accurate prediction of the wind speed is of great significance for improving the accuracy of wind power prediction and the stable operation of the power grid. The precise characterization of the prediction model residuals is a prerequisite to achieve accurate prediction of the wind speed series in the wind farm. This paper proposes an ultra-short-term wind speed hybrid forecasting model based on the probability of time series residuals. First, the wind speed is decomposed into components with different frequency characteristics based on the optimized variational modal decomposition. Then, a deterministic prediction model is constructed for the linear components with regular changes in the wind speed components through the time series model. For the fitting residual components, the conditional kernel density estimation is used to establish a probability forecasting model. Then based on the superposition of the recursive results of the two models the wind speed prediction value is formed. On this basis, in view of the problem that the residual conditional probability of each component cannot directly represent the original wind speed probability forecasting result, this paper proposes a probability generation based on the simulation to realize the wind speed probability forecasting. Finally, taking the operating data of a wind farm in Northeast China as an example, it is verified that the proposed method has high forecasting accuracy. While ensuring the reliability, the proposed method has a very low prediction interval width, which reduces the uncertainty of the probability forecasting.