Meticulous Study On Wind Forecasting Over The Complex Terrains

Accurate wind forecasting over typical terrains in China was conducive to the development of the wind energy industry and the sustainable development of China’s economy. Jiangsu and Inner Mongolia were selected respectively as typical representatives of coastal beaches and inland plateaus complex terrains in this paper. Comparative tests were set based on the mesoscale model WRF V3.3.1(Weather Research and Forecasting Model) with two different boundary layer parameterization schemes (YSU/MRF). At last, the WRF/Noah/YSU were selected as wind forecasting system over typical terrains in China. Using this system, the48h wind rolling forecasting were conducted separately in two study areas in2010with1km horizontal resolution,10min time resolution. The WRF/Noah/YSU wind forecasting scheme was proved to be more accurately in the aspect of forecasting wind field characteristics by further examining the power spectrum, the wind roses of direction and speed, seasonal and diurnal variation. The distribution of power spectral, the characteristics of dominant wind speed and direction as well as diurnal and seasonal variation of forecasting was basically consistent with observation in the masts. The forecasting results were better in the spring and autumn than in the summer, and the diurnal variation of forecasting varied due to terrains with no uniform rules.Besides that, in this paper, some statistical forecasting methods from the perspective of a single statistical forecasting method and combined forecasting models are introduced, such as time-series, kalman filtering, neural network, support vector machine, wavelet analysis, wavelet-support vector machine and so on. Compare their advantages and disadvantages comprehensively to use efficiently the advantage of each method in wind forecasting. Through reviewing and summarizing the existing research results, it is found that in the single statistical forecasting methods, the traditional statistical forecasting method is suitable for handling simple wind forecasting with relatively low forecasting accuracy. Artificial intelligence is superior to the traditional methods for solving complex wind prediction problems with higher forecasting accuracy. Combination of statistical forecasting methods can overcome the limitations of the single statistical methods and become one important way to improve the accuracy of wind forecasting.At last, in this paper, Inner Mongolia was selected as the representative of complex mountainous terrains with abundant wind energy resources. Based on the numerical prediction results of four masts in study areas in2010, BP neural network (BP-ANN), least squares support vector machine method (LS-SVM) were selected to revise short-term wind forecasting with l0min time resolution; Persistence method-BP neural network method, Persistence method-least squares support vector machine method were selected respectively to revise ultra-short-term (P6-BP, P6-SVM) and nowcasting (P1-BP, P1-SVM) wind speed with lOmin time resolution and compare their difference. By examining overall forecasting results, the wind roses of speed, seasonal and diurnal variation, it is found that the wind speed after revised was more close to the observations and the revision was remarkable. In addition, the revision of nowcasting was better than ultra-short-term than short-term forecasting. By examining the wind speed at different levels, all revising models existed larger error when the wind speed was larger and smaller error when the wind speed was smaller. The revision of intermediate stage (8-14m/s) was best of all wind speed stage.