Abstract:Accurate wind speed forecasts ensure the timely dispatch of power,thereby improving the economic effectiveness of wind power.The traditional methods of wind speed prediction focus on statistical models,but these are unable to satisfy requirements in terms of precision and time.Numerical models have become more prevalent in wind speed forecasting in recent years.Accordingly,in order to investigate simulation accuracy,error characteristics and time effectiveness for wind speed simulation of a numerical model over a complex underlying surface,one-month long wind speeds at the heights of 10 m,30 m,50 m and 70 m on a wind farm in Liuao (coastal area) and Jiucaiping (mountain area) are respectively simulated using the WRF model.The main results and discussion points are as follows:
(1)The comparison of simulated and observed wind speeds at different heights indicates that the WRF model performs well in simulating the wind speed over a complex underlying surface,the simulated wind speed trends are in good agreement with the observations in both coastal and mountain areas,and the synoptic-scale variations are also well reflected in the simulation.However,the fluctuation caused by the local circulation and turbulence is hard to capture in the wind speed simulation.Besides,statistical assessments using correlation coefficients,relative error,root-mean-square error(RMSE) and index of agreement show that the accuracy of wind speed simulation does not enhance with increased height in coastal areas,but improves significantly in mountain areas.
(2)The error characteristics between simulations and observations are different in coastal and mountain areas.Compared with the real landforms,the small island in coastal areas does not appear in the model's static terrestrial data.As a result,the frictional effect of the underlying island is neglected in the simulation,and the average simulated wind speeds are overestimated in coastal areas.Nevertheless,the elevation of the wind tower in mountain areas decreases in the static terrestrial data due to the weakening of the actual mountain slope,and thus the average simulated wind speeds are underestimated in mountain areas.
(3)The normalized RMSE of wind speeds is analyzed in different directions.The normalized RMSE increases significantly due to the relatively complex underlying surface in the case of land breezes,and decreases due to the homogeneous underlying surface in the case of sea breezes in coastal areas.Conversely,the normalized RMSE distribution is not obvious in mountain areas.The obvious error distribution characteristics provide a direction for further error correction of wind speed forecasting on wind farms in coastal areas.
(4)To forecast the wind speed on a scale of a few hundred kilometers surrounding the wind farm,parallel computation of the WRF model using a server with 12-core processors is sufficient to meet the time effectiveness of 48-h short-term forecasts.However,only increasing the grid resolution is not necessary to improve the accuracy of the simulation.Therefore,in the practical application of wind speed forecasting on a wind farm,it is important to set an appropriate grid resolution for the balance between simulation accuracy and time effectiveness.More accurate terrestrial data should be introduced to improve the precision.