WRF/CALMET/BP预报系统对内陆山区风电场的一次滚动预报检验

A numcial testment of WRF/CALMET/BP forecast system for wind energy in inland mountains

利用以中尺度数值模式WRF/CALMET作为风电场预报系统的动力模块,及BP神经网络法(BP-ANN)作为风电场预报系统的统计订正方法,对重庆市齐跃山风电场进行了一次时间分辨率为5 min的24 h风速、风功率的滚动预报试验,探讨了适用于中国典型内陆山区的风电场预报系统。结果显示:以WRF/CALMET/BP组成的动力—统计预报系统能够较好地模拟出内陆山区的风场特征,系统对正午至傍晚时段的风速预报准确率较高。WRF/CALMET动力模式对于风速中心振幅的模拟能力较好,经过BP神经网络订正后,模拟结果会趋于均值。不同风速段中,模式对低风速段(3~8 m·s^(-1))的预报效果较好,BP神经网络对中风速段(8~14 m·s^(-1))预报结果的订正效果最明显。

A wind energy farm forecast system based on the mesoscale numerical model WRF/CALMET and the post statistical BP Neural Network Method(BP-ANN),was used to operate the rolling prediction test on 24 h wind speed and wind power at 5-minute intervals of Qiyueshan wind farm.Results indicate that dynamic-statistical forecast system of WRF/CALMET/BP simulates the wind field characteristics in inland mountainous well,whose accracy rate is highest from noon to sunset.The numerical model WRF/CALMET simulates central amplitude well,but after statistical revision by BP-ANN,the results tend to distribute near mean value.Meanwhile,the forecast accuracy in low-level wind speed(3-8 m·s^(-1))by model is better,and the effort of statistical revision by BP-ANN is more obvious in mid-level wind speed(8-14 m·s^(-1)).