1) Understanding wind energy characteristics, analysis wind energy probability distribution and predicting wind energy resources are key link in ensuring the safety of wind power generation. According to fujian coastal wind tower real-time measurement data, the analysis of daily, monthly and year average wind speed data can effectively describe average wind conditions, using the least squares method calculate the shape parameter and scale parameter of weibull distribution and fit wind tower each layer height wind speed frequency distribution, the fitting curve can describe the coastal of fujian province wind tower wind speed distribution in each layer, and pass0.1x square test. This result suggest near sea surface wind speed probability characteristic can be described by weibull distribution.2) Collect MM6numerical forecast products of surface wind speed from June2010to December provided by CMA, interpolate the time interval form3h to1h linearly, and interpolate the grid wind data to the position of the wind tower through the distance weighting method, finally vertical interpolate the wind data to height of wind towers sonde according to surface layer wind speed exponent profile, reach the space and time synchronization of the observation and MM5numerical forecast product data. Establish the BP neural network dynamic correct the simulation of MM5numerical forecast product wind speed. Compare the corrected wind speed Vmm with real-time observation wind speed Vt,it is found that the Vmm mean absolute error is reduced by10%~20%, and the correlation coefficient between Vmm and real-time wind speed has increased significantly. These show that, through training the BP neural network can correct MM5wind speed simulation error effectively.3) Select different boundary layer, the surface layer and land surface process parametrization in the WRF model, design four different physical processes parameterization combined scheme simulate wind speed in a fujian coastal wind tower station during January1toll and July1to11in2010, compare the hourly simulation results with the wind tower real-time observation data to seeking the best parameterization scheme. Through analysis and comparison, the scheme2which using the MYJ boundary layer scheme, Monin Obukhov surface layer sheme and Noah land process has the best simulation result. Using this scheme to stimulate wind speed in January2010and July and compare to observation according to the different wind speed level respectively, the results show that WRF (scheme2) stimulation mean relative error for6to15m/s wind speed is about20%, which can meet the demand of wind power prediction accuracy; but the0-6m/s wind speed simulation error is relatively large, which may due to model the lack of model terrain resolution and the special wind towers location that on the border of sea land, so lower wind speed is vulnerable to ground disturbance and land and sea breeze. |