Wind Power Forecasting has important significance to optimize the power dispatching plan, and promote wind power acceptance. By comparing with the historical data of load concerning the daily fluctuation characteristics and the probability distribution, the volatility of the day data of wind power has no obvious law to follow on year-on-year week-on-week basis, and its statistics law approximately meets Weibull distribution, which undoubtedly increases the difficulty of wind power prediction technology. An identification method of the intimate sample based on entropy-weight index and correlation sorting is proposed, in order to mine useful information fromSaSmassSofShistorical data for short-term wind power forecasting model. Through case study of the northern province of real data, it show that, compared with other common prediction algorithms, the improved method proposed in this paper has some advantages in improving the accuracy and the efficiency of short-term wind power forecasting.