The studied wind farm is located in the middle of Inner Mongolia,China. The real-time gradient observation of the wind tower and the surrounding automatic meteorological station for the wind farm is taken as the benchmark.Using the observed wind speed and wind direction（one year's data） at the wind tower located in the study area , and wind speed and wind power at turbines'hub height for the same period,The quality control method for wind measurement data on turbine hub height is studied,analyzing the environment and the Characteristics of wind flow change. According to the different seasons and wind direction (NW, SW, NE, SE), comprehensive analyses of spatio-temporal correlation and deviation of wind tower's and turbines'wind speed are carried out. The partitions have been done in A according to the different seasons and different wind direction. The partitions are made of the turbines whose wind speed are characterized by high correlation (>0.8). The characteristics of wind speed of wind tower and the partitions that are affected by turbulence, wake and wake superposition are studied. Using wind tower's and turbines' wind speed in their partitions, the fitting equations are established.Wind profiles, correlation ratios, autoregressive differential movement and other methods are selected to study the quality control of assimilation application of wind measurement data from wind turbine hub in wind farm, and the application control model of high correlation zoning classification wind speed analysis is established.Reduce the temporal and spatial consistency bias of short time wind speed and wind speed in the Wind Turbine Partition for the Wind Farm.The learning model based neural network is applied to the short-term turbines' wind speed prediction bias correction of numerical prediction.At the same time, it will be applied to real-time data assimilation and post-processing of high resolution wind energy numerical weather prediction model,and high-precision wind power prediction and operation evaluation, etc.