The Research On Revised Methods Of Numerical Weather Prediction Of Wind Velocity In Wind Farm

Using the MM5material released by the Central Meteorological station and1year observed data in wind farm in pingtan County and zhangpu County of fujian province, we analysed the wind speed characteristics and error characteristics, calculated with fall scale interpolation method, we learn that:1, The local wind velocity and wind power distribution characteristics, and there is a scope of timing delays between the diurnal variation characteristics near the sea delay and inland. The wind speed distribution mainly concentrates in northeast and north-northeast, the wind is weak in other directions, in southwest direction it account for roughly10-15%. Weibull distribution fitting effect of wind speed is good, belongs to the gently type distribution, which is beneficial to the wind generator installation and operation, commissioning and making development plan.2, The wind speed by micaps system is in low temporal and spatial resolution, so the mean absolute error (MAE,6h forecast) by interpolation is almost above2m/s, the errors caused by interpolation methods and random is greater than model calculation error itself, doing separation calculation base on different causes of error provides a good thought for further improveing predicting accuracy.3, By calculating and analysing autocorrelation coefficient of the MM5forecast wind speed error, The relationship close to the time appear to be highly relevant features, by time delayed of sample, the linear relationships decreased. In six hours, autocorrelation can almost reach significant related standards.Diurnal variation of MM5wind speed error is relatively complex, usually have two to three base area and two high value area, overall, error in the1st tower is significantly greater than error of the2nd tower. Diurnal variation of error in summer and autumn is significant, winter next, diurnal variation of error in spring is relatively small, more stable.4, Effect of he mean average error direct correction method is general, the forecast effect indexes were not improved greatly. According to the former method, the effect of systematic correction method raised, and the mean absolute error is improved, but still be above2m/s, the prediction error is still relatively large and can not reach the requirement of engineering design.5, Little error correction method adopts the dynamic revised method, establishing the regression equation by using the actual forecasting errors and related factors several times before. Use stepwise regression method to choose the factors at the0.05significance level. According to the actual forecast effect, it is considered that the errors two times before and wind speed standard deviation factors are established0-6h corrections equation, and the results show that the method can significantly improve short-term prediction precision while the mean absolute error of the1st tower in each season is reduced respectively by0.68m/s,0.75m/s,1.11m/s,0.95m/s, decreased about31%-42%, the2nd tower reduced by1.61m/s,0.84m/s,0.78m/s,1.55m/s, decreased about34%-54%.6, Take the15days material in December2010as test period, the results show that the proposed method in the revision of the new sample has significant effect, generalization ability is strong. But considering the coarse grids and low precision of MM5numerical prediction product by the Central Meteorological station, we hold the view that this method combined with higher resolution numerical prediction product may has better forecast effect.