Abstract: Predicting the wind power output in the next few hours is crucial for the safe and economic operation of the power system as well as the participation of wind power in the electricity market. Due to the strong randomness and non-stationarity of the wind power sequence, it is necessary to characterize and model its uncertainty. Moreover, factors such as equipment aging, blade contamination, and changes in the wind farm environment can affect the output characteristics of wind turbines, thereby making the parameters of the wind power prediction model time-varying and increasing the difficulty of wind power prediction. This paper proposes a short-term wind power probability prediction method based on online Gaussian process. First, the Gaussian process regression model is used to model the wind power prediction problem. Then, the complexity of Gaussian process calculation is reduced by combining structured kernel interpolation with the Woodbury identity, which enables fast Gaussian process solving.Finally, the method of block caching and updating is adopted to realize the real-time online updating of parameters and hyperparameters for Gaussian process model. The wind power generation data released by the 2014 Global Energy Forecasting Competition is used to validate the proposed algorithm. The results show that the proposed algorithm has good prediction performance and adaptability to effectively deal with the problem of time-varying model parameters.