In order to improve the accuracy of short-term wind power prediction, a short-term wind power prediction method based on UP-DOWN-features of wind speed is presented. By analyzing the change of output power caused by UP-DOWN-features, wind speed is labeled by the features in every moment to increase the training dimensions and then the prediction accuracy is improved. The data from a wind farm in Shanghai from September 2014 to September 2015 is used. By comparing the prediction results of Least Squares Support Vector Machine (LSSVM), Extreme Learning Machine (ELM), Genetic Algorithms-BP Neural Networks (GA-BP), the prediction accuracy of the model which has added the UP-DOWN-features will be enhanced, and experiments show that it is suitable for short-term wind power prediction.