摘要
建立了基于BP神经网络的PM_(2.5)质量浓度预报模型,对广州市5个监测点2012年6月—2013年5月的PM_(2.5)质量浓度日均值进行预报,分析了总体预报误差、不同风速和降雨量下的预报误差,以及天气预报误差对PM_(2.5)质量浓度预报误差的影响。结果表明,BP神经网络模型对5个站点的PM_(2.5)预报结果稳定,平均相对误差为29.71%。在有利于PM_(2.5)扩散的气象条件下预报误差较大,风速较大时与风速较小时预报误差的差异高达15%,而不同降雨量情况下的预报误差较相近。修正天气预报后,各站点的预报误差平均降低了4.67%。这表明可从空气质量数据质量等方面入手改进模型。
This paper would like to present our research results of the effects of meteorological parameters on the daily forecasting model of PM2.5 concentration based on the BP Neural Network.The said model is supposed to be used for the five monitoring stations under the Guangzhou administrative region,that is,the stations of Baogao,Chisha,Tianhecheng,Luhu and Tianhu.Seeing the actual needs of the forecasting mission,we have analyzed the forecasting results under different polluted meteorological conditions by choosing the daily average PM_（2.5） concentration of the national standard.Among the manifold meteorological parameters that may impact the PM_（2.5） concentration,we have selected the wind speed and the rainfall as the key ones among them.In hoping to overcome the inaccuracy of the weather forecasting concerning that of PM_（2.5） concentration,we have been keeping the forecasting results and comparing the recorded meteorological data with those of the predicted ones.The results show that the mean relative errors in the five monitoring stations turn out to be from 27.77%to 32.91%,which is stable and reliable.And the mean relative error in the five monitoring stations works out at about29.71%.The forecasting results tend to be better if the weather may not be conducive to the diffusion of PM2.5.What is more,the forecasting results under different rainfall grades turned out the same with the mean relative error under different rainfall grades being within2.51%,though the mean relative error on the condition of high wind may turn out a bit higher than under the low wind by a cap of14.65%.When the error of weather forecast has been rectified by a weighting factor,the mean relative error can be made equal to25.04%.The reduced results of the error of the said five stations stands from 2.75%to 10.15%,with the average error being4.67%.Thus,it can be concluded that these results show the different forecasting errors under different weather conditions with the average forecasting error caused by the inaccuracy of w
出处
《安全与环境学报》
CAS
CSCD
北大核心
2015年第6期324-328,共5页
Journal of Safety and Environment
基金
国家自然科学基金项目(51108471)
关键词
环境学
PM2.5日均值预报
BP神经网络
气象参数
预报误差
environmentalology
PM（2.5） daily concentration forecasting
BP neural network
meteorological parameters
forecasting error
