气溶胶散射吸湿增长因子的BP神经网络模型  

A BP NEURAL NETWORK MODEL OF AEROSOL SCATTERING HYGROSCOPIC GROWTH FACTOR

作　　者：佟景哲 倪长健[1,2] 米家媛 蒋梦姣 张莹 石荞语[3] TONG Jingzhe;NI Changjian;MI Jiayuan;JIANG Mengjiao;ZHANG Ying;SHI Qiaoyu(College of Atmospheric Science,Chengdu University of Information Technology,Chengdu 610225,China;Chengdu Plain Urban Meteorology and Environment Observation and Research Station of Sichuan Province,Chengdu 610225,China;Institute of Plateau Meteorology,China Meteorology of Administration,Chengdu 610072,China)

机构地区：[1]成都信息工程大学大气科学学院,成都610225 [2]成都平原城市气象与环境四川省野外科学观测研究站,成都610225 [3]中国气象局成都高原气象研究所,成都610072

出　　处：《环境工程》2023年第7期131-137,165,共8页Environmental Engineering

基　　金：四川省科技厅应用基础研发项目(2021YJ0314);国家重点研发计划项目(2018YFC0214004,2018YFC1506006)。

摘　　要：气溶胶组分、结构以及形态的复杂性对高湿条件下气溶胶散射吸湿增长因子统计模型的适用性提出了挑战。基于成都市2017年10—12月浊度计和黑碳仪的逐时观测资料,结合同时次的环境气象监测数据,利用“光学综合法”计算气溶胶散射吸湿增长因子。以相对湿度(RH)、C_(BC)、C_(BC)/C(PM_(2.5))、CPM_(1)/C(PM_(2.5))以及C(PM_(2.5))/C(PM_(1)0)作为输入因子[C_(BC)、CPM_(1)、C(PM_(2.5))、C(PM_(1)0)分别为黑碳(BC)、PM_(1)、PM_(2.5)、PM_(1)0的质量浓度],构建了气溶胶散射吸湿增长因子的BP神经网络模型。结果表明:气溶胶散射吸湿增长因子BP神经网络模型、多变量GAM模型、双变量模型以及单变量二次多项式模型的判决系数(R^(2))分别为0.870、0.792、0.744和0.650,其中高湿条件(RH>85%)下模拟值的判决系数(R^(2))分别为0.749、0.685、0.638和0.538。多模型对比表明,气溶胶散射吸湿增长因子BP神经网络模型的模拟效果最优,显著降低了统计模型在高湿条件(RH>85%)下的模拟误差。The complexity of aerosol components,structure and morphology challenges the applicability of statistical models for aerosol scattering hygroscopic growth factor under high humidity conditions.Based on the hourly observational data of nephelometer and aethalometer,as well as the simultaneous monitored data of environmental meteorology from October to December,2017 in Chengdu,the aerosol scattering hygroscopic growth factor was calculated by optical synthesis method.The relative humidity(RH),C_(BC),C_(BC)/CPM_(2.5),CPM_(1)/CPM_(2.5) and CPM_(2.5)/CPM_(1)0 were used as input factors(C_(BC),CPM_(1),CPM_(2.5) and CPM_(1)0 representing mass concentrations of BC,PM_(1),PM_(2.5) and PM_(1)0 respectively),and a BP neural network model of aerosol scattering hygroscopic growth factor was proposed.The comparison results of multiple models showed that:the corresponding determination coefficient(R^(2))of the univariate quadratic polynomial model,bivariate model,multivariate GAM model and BP neural network model for the aerosol hygroscopic growth factor were 0.650,0.744,0.792 and 0.870 respectively,and the corresponding determination coefficient(R^(2))for the simulated values under high humidity conditions of RH>85%were 0.538,0.638,0.685 and 0.749,respectively.The BP neural network model of aerosol scattering hygroscopic growth factor achieved the best fitting effect and the simulation error of aerosol scattering hygroscopic growth factor under high humidity condition (RH>85%) was significantly reduced.

关 键 词：气溶胶 散射吸湿增长因子 人工神经网络 BP模型 成都 

分 类 号：X513[环境科学与工程—环境工程]