An integrated forecasting method of short-term wind power is presented for improving prediction accuracy and shortening the model training time. Based on the characteristic of wind power curve and wind speed frequency, the wind speed is divided into high, medium and low three segments, and each wind power characteristic is analyzed. As the predicted power shows larger fluctuated statuses in segments of high and low wind speed, so Least Squares Support Vector Machine is used to achieve better prediction accuracy. Much more data can be accessed in the medium segment, and there is an obvious physical relationship between wind speed and power, so Gaussian Model is used under this sort of circumstance. At the same time, the level table of wind and power is used to revise the predicted power in each section to ensure the stability of the algorithm. The rationality of Gaussian model and selection of algorithm in high, medium and low segment is verified by using the historical data of a wind farm of Shanghai in 2014. The simulated result compared with LSSVM’S shows that the proposed algorithm can not only improve the prediction accuracy, but also shorten the model training time. It can be well used to predict short-term wind power in real-time.