径向基函数网络在大气颗粒物来源解析中的应用  

Analysis of Source Contributions to the Ambient Aerosol Sample by Radial Basis Function Neural Network

作　　者：胡焱弟[1] 李新欣[1] 白志鹏[1] 冯银厂[1] 赵玉杰[1] 吴建会[1] 

机构地区：[1]南开大学环境科学与工程学院,国家环境保护城市空气颗粒物污染防治重点实验室,天津300071

出　　处：《过程工程学报》2006年第z2期86-90,共5页The Chinese Journal of Process Engineering

基　　金：国家自然科学基金资助项目(编号:20307006);南开大学2005年度本科生创新科研"百项工程"

摘　　要：将径向基函数网络(Radial Basis Function Network,RBFN)应用于城市环境颗粒物来源解析工作.模拟数据计算的解析结果表明:RBFN可以实现对多源(14个可能源,其中13个为有效源)的解析,在5%～15%的源和受体测量误差的情况下,对于分担率大于15%的主要源,其解析结果与真实值的相对误差均不高于5%;对于分担率大于5%的源,其解析相对误差均低于15%.RBF网络可以很好地识别无效源.因此,在充分掌握可能污染源成分谱信息的基础上,该方法具有源解析应用潜力.A radial-basis function artificial neural network(RBFNN) is proposed as a receptor modeling method for solving ambient airborne particulate source apportionment problems while multiple possible sources exist.The results show that RBF leads to satisfactory results.With 5%～15% relative errors for the source profiles and ambient concentrations,the relative errors between the calculated source contributions and the truth are no higher than 5% for the sources with loadings higher than 15%,and are lower than 15% for the sources with loadings higher than 5%.Moreover,when training set containing all of the possible emission sources,some of them are not active,the RBF network is able to identify the inactive sources quite well.So we come to the conclusion that with adequate source profile information of all possible sources,the RBF network is promising method to be used in source apportionment.

关 键 词：大气颗粒物 源解析 径向基函数网络 模拟数据 

分 类 号：X51[环境科学与工程—环境工程]