基于滚动极值处理的BP神经网络方法的WRF模式预报风速订正  

Application of BP Neural Network Based on Rolling Extremal Management in Revising Wind Speed Forecasting of WRF Model

作　　者：黄凤新[1,2] 刘寿东[1] 祝赢[3] 柳艳香[2] 

机构地区：[1]南京信息工程大学应用气象学院,南京210044 [2]中国气象局公共气象服务中心,北京100081 [3]武汉中心气象台,武汉430074

出　　处：《科学技术与工程》2013年第7期1768-1772,共5页Science Technology and Engineering

基　　金：公益性行业(气象)科研专项(GYHY201206054);(2010DFA62830);"长江学者和创新团队发展计划"共同资助

摘　　要：为解决风速预报准确性问题,提高风能利用率,应用滚动的BP神经网络方法结合对极值的处理,对我国某风电场2007年4—11月的WRF模式预报风速结果进行了滚动订正研究。结果表明,经过1 h、12 h、24 h时间步长的滚动极值BP神经网络方法订正后,风电机组轮毂高度70 m处订正风速的相对均方根误差平均值分别减少16.59%、12.11%和11.61%,相对平均绝对误差平均值分别减少13.81%、11.09%和10.37%;且以时间步长为1h的滚动极值处理的BP神经网络订正效果最好,预报风速更趋近于实测风速,预报精度明显提高。In order to improve the utilization of wind energy, the accuracy of wind speed forecasting should be raised. The wind speed of a wind farm in China forecasted by WRF model from april to november in 2007 is re- vised, by using the rolling BP neural network method combined with the management of the extremum. The results showed that, after being revised with the time step of 1 h, 12 h, 24 h, the average rRMSE （relative root mean square error） of the revised wind speed had reduced by 16. 59%, 12. 11% and 11.61% respectively at the height of 70 m of wind turbine wheel hub height, and the average rMAE （relative mean absolute error） had decreased of 13.81%, 11.09% and 10. 37%. Furthermore, the revised result with the time step of 1 h was closer to the meas- ured wind speed , the predicting precise can be improve obviously.

关 键 词：滚动极值处理 BP神经网络 WRF模式 风速订正 

分 类 号：TK81[动力工程及工程热物理—流体机械及工程]