两种风速短临预报方法对比研究

Short-term wind speed forecasting study based on two methods

利用BP神经网络法和最小二乘法,对不同地形条件下的4个测站的10 s量级和15 min量级平均风速进行短临预报实验。研究发现,最小二乘法预报误差小,满足预报误差小于35%的日数比较大。无论是10 s量级预报,还是15 min量级预报,对于风速较大的01号站和04号站,最小二乘法优于BP神经网络法;对于风速较小的02号站和03号站,两种预报方法的预报效果相近;在10 s量级和15 min量级的风速短临预报方面,算法复杂的BP神经网络法并无明显优势。因此,在选取预报方法前,应结合预报方法本身的特征,充分考虑预报方法对地形、地貌和气候特征以及预报时效的适应性,最好对几个备选方法进行预报效果比对。

There two groups of experiments of 10-seceond＇s magnitude and 15-minute＇s magnitude wind speed short-term forecasting have been done by using BP neural network method and least square regression method on 4 stations with different topography. The study found that both magnitude of 10-seceond and 15-minute forecasting experiments show：（1）The method of least square regression is better than BP neural network for 01# and 04# station with larger wind speed. The error is less and forecast satisfaction ratio is larger of least square regression method than that of BP neural network.（2） For 02# and 03# station with smaller wind speed, the forecast effect is similar between these two methods. In other words, there is no advantage of BP neural network method, which with complex algorithm, in very short-term wind speed forecasting, such as magnitude of 10-second or 15-minute forecasting. Therefore,the algorithm characteristics, the topography, geomorphological,climatic characteristics and period of validity should be full consideration, and the comparison test of forecasting should be performed before the method be selected.