With the large-scale wind power connected into the power systems,the intermittency and randomness of wind power have a great impact on the stability of the power systems.Therefore,the accurate prediction of wind power has become one of the most important ways to solve this problem.In this paper,considering the timing memory characteristics of long short-term memory (LSTM) network,by combining wavelet decomposition and LSTM network,an ultra-short-term probability prediction model for wind power based on wavelet-LSTM network is proposed.Firstly,wavelet decomposition is used to smooth the sequence of the original time sequence.Then the LSTM network prediction model for the sequence samples is developed.By using the maximum likelihood estimate method,the Gaussian distribution function of prediction error can be estimated.Thus the probability prediction of wind power in the future 4 hours could be realized.Finally,based on the wind farm data in Northeast China,simulation results show that wavelet decomposition with deep learning method can improve the accuracy of prediction.The interval reliability of probability prediction is also improved.