The Research On The Wind Power Prediction On Yangmei Mountain Wind Farm Based On The WRF Weather Forecast Mode And SVM Statistical Regression Approach

Based on the WRF weather forecast mode and SVM statistical regression approach, the research on the wind power prediction on Yangmei mountain wind farm located at typical low latitude plateau was carried out by this thesis. By selecting July and October of2012, January and April of2014as weather prediction target months and by employing three types of surface layer approximation schemes (namely MM5approximation scheme, ETA approximation scheme and PX approximation scheme), the meteorological element farm simulation experiments aiming at a3-day-prediction with the time resolution of15min were performed. After that, the simulation results were extracted and the wind velocity, which is the prime factor of wind power, was analyzed in a detailed manner. Lastly, by incorporating meteorological factors, such as wind velocity, direction, temperature, air pressure and the actual wind power, into SVM-based regression equation,72-hour wind power prediction on Yangmei mountain wind farm was finally carried out. The influence that the training data in SVM-based regression exerted on predication results was also analyzed. The principal conclusions are listed as follows:(1) On the aspect of wind velocity simulation, the three surface-layer schemes of WRF mode are capable on the simulation of the changing trend of wind velocity. One thing need to be noticed is that the wind velocity amplitude simulated by WRF mode is bigger than the actual wind velocity. Namely, if the actual wind velocity increased, the WRF-simulated wind velocity increased at a larger amount; if the actual wind velocity diminished, the WRF-simulated wind velocity diminished more.The best wind velocity simulation was the prediction on the first day of a3-day prediction in January,2013made by MM5approximation scheme with the correlation of0.81. The minimum root-mean-square error (RMSE) and minimum mean-absolute error (MAE) of wind velocity were occurred on the second day of a3-day prediction in October,2012made by MM5approximation scheme with the figures estimated at1.94m/s and1.491m/s. The best correlation of wind velocity simulation was reached on January,2013while the worst on July,2013. The minimum RMSE and minimum MAE were occurred on October,2012in comparison to the maximum of that on April,2013. Among the three schemes of WRF simulation, MM5approximation worked the best. On the performances of RMSE and MAE, the ascending ranking ended up at the sequence of ETA, MM5and PX, respectively.(2) All of the three WRF approximation schemes had passed the significance tests on wind power prediction with the effectiveness ranked at, from high to low, MM5, ETA and PX, respectively. On RMSE and MAE evaluation, MM5and ETA performed far better than PX, which made MM5and ETA capable of wind power prediction task while failed PX as a qualified solution. The best correlation of wind power predication was obtained on January,2013while the worst on April,2013. The Minimum RMSE and MAE was on July,2012while the maximum on April,2013. When detailed into the comparison among each month and each approximation scheme, it turned out that, in terms of correlation, the best prediction was done on October,2012by ETA approximation while the worst was on the same month of2012by PX approximation and, in terms of RMSE and MAE, the minimum error was done on July,2012by MM5approximation while the maximum on April,2013by PX.In the process of SVM statistical regression, the prediction results were influenced by the selection of training data.(3) On the basis of WRF mode and SVM statistical regression approach, combined with the all above-mentioned comparisons, it is concluded that MM5surface layer approximation scheme provided the best simulation for the wind power prediction on Yangmei mountain wind farm.