This paper studies short-term prediction of wind power generation capacity by means of wavelet transform and relevance vector machine（RVM）. Firstly

the original wind speed signal sequence is decomposed into a series of sub-sequences with different frequency by wavelet transform

and the low-and high-frequency components in the wind speed signal are extracted. Then

RVM is applied to these sub-sequences for respective prediction considering the data of wind direction and temperature

and final wind power prediction is hence obtained through reconstruction of the respective predictions. Finally

the proposed prediction method is applied to a certain wind farm

and simulation results show that wavelet transform can help to understand the variation of wind speed

and the combination of wavelet transform with RVM

which has better learning ability

can effectively improve the prediction accuracy.