基于支持向量机和树叶磁学的大气颗粒物浓度模拟  

Simulation of Atmospheric Particle Concentration based on Support Vector Machine(SVM) Method and Leaf Magnetics

作　　者：许悦 刘雪梅 周梦帆 李慧明[1] 钱新[1] XU Yue;LIU Xue-mei;ZHOU Meng-fan;LI Hui-ming;QIAN Xin(State Key Laboratory of Pollution Control & Resources Reuse, School ofthe Environment, Nanjing University, Nanjing 210023, China)

机构地区：[1]南京大学环境学院污染控制与资源化研究国家重点实验室

出　　处：《四川环境》2019年第3期88-95,共8页Sichuan Environment

基　　金：国家自然科学基金项目(41771533);江苏省自然科学基金面上项目(BK20171339)

摘　　要：在南京市仙林地区采集桂花、雪松及女贞树叶,测试其磁学参数,分析大气颗粒物浓度与树叶磁学参数、气象数据的相互关系,使用支持向量机方法构建大气颗粒物磁学评估模型。结果显示,雪松树叶的质量磁化率(χ)、饱和等温剩磁(SIRM)和非磁滞剩磁(ARM)值最高。3种树叶在夏季和秋季富集了相对较多的细磁畴磁性颗粒物,而在春季和冬季富集了相对较多的粗磁畴磁性颗粒物。PM10和PM2.5浓度与风速、温度和湿度之间显著负相关,与大气压显著正相关;与3种树叶的磁学参数显著正相关,且与χ和SIRM的相关性高于ARM,其中与桂花树叶χ、SIRM的相关系数均高于0.560。将“树叶磁学参数+气象数据”作为输入因子的支持向量机模型对PM2.5浓度的模拟效果较好,将“树叶磁学参数”作为输入因子的模型对PM10浓度的模拟效果较好,模型误差都在可接受范围内。Samples of Osmanthus fragrans Lour, Ceder deodara G.Don and Ligustrum lucidum Ait were collected in Xianlin district of Nanjing. The magnetic parameters of tree leaf samples were measured. The correlations between concentrations of atmospheric particulate matter and magnetic parameter of leaves, as well as meteorological data were analyzed. The support vector machine (SVM) method was used to construct the magnetic evaluation model of atmospheric particulate matter. Results showed that the value of mass magnetization rate (χ), saturated isothermal remanence (SIRM) and anhysteretic remanent magnetization (ARM) of leaves of cedar were the highest. The three leaves accumulated relatively more fine-magnetic domain magnetic particles in summer and autumn, whereas relatively more coarse-magnetic domain magnetic particles in spring and winter. PM 10 and PM 2.5 concentrations showed significant negative correlation with wind speed, temperature, and humidity, and significant positive correlation with atmospheric pressure. The magnetic parameters of three species of leaves had positively correlations with PM 10 and PM 2.5 concentrations (p<0.05). The correlation coefficients (r) of magnetic parameters of tree leaves were higher with χ and SIRM than with ARM.The correlation coefficient (r) of magnetic parameters of Osmanthus fragrans Lour leaves with PM 10 and PM 2.5 concentrations were both higher than 0.560. Taking “leaf magnetic parameters and meteorological data” as the input factors, the simulation effect of SVM models on PM 2.5 concentration was better, while the simulation effect of taking only “leaf magnetism parameter” as the input factor on PM 10 concentration was better. All the model errors are within the acceptable range.

关 键 词：大气颗粒物 支持向量机 磁学参数 树叶 南京市 

分 类 号：X513[环境科学与工程—环境工程]