光谱学与光谱分析, 2018, 38 (2): 643, 网络出版: 2018-03-14
基于太赫兹光谱-统计分析的大气PM2.5监测研究
Terahertz-Statistics-Dependent Approach for Monitoring PM2.5 in Air
摘要
应用太赫兹技术对大气中动力学直径小于25 μm(PM25)的细颗粒物进行了定量研究。 PM25质量和太赫兹吸光度之间存在线性关系, 相关系数为086。 应用主成分分析的方法, 可证明随着PM25质量的增加, 与吸收系数存在相似的趋势。 为了提高预测精度, 采用偏最小二乘, 支持向量机和反向传播人工神经网络对PM25进行定量研究。 与单一的线性模型相比, 统计模型具有较大的预测相关性和较小的误差。 对于神经网络模型, 训练集与预测集的相关系数和均方根误差分别达到0999和0016 mg, 0912和0207 mg。 因此, THz技术和统计学方法的结合可提供较高精度的预测, 作为一种监测PM25的有效手段。
Abstract
The study presented a quantitative investigation in monitoring the fine particulate matter with the aerodynamic diameters less than 25 μm (PM25) in air with terahertz (THz) technique. Absorption bands were observed and linear relations were obtained between PM25 mass and absorbance at selected frequencies with correlation coefficients R of ~0.86. The utilization of principal component analysis (PCA) proved the similar absorption trend in the entire range with increasing PM25 mass. In order to improve the prediction precision, we employed other three statistical methods including partial least square (PLS), support vector machine (SVM) and back propagation artificial neural network (BPANN) for the quantification of PM25. Compared with single linear models, statistical models had larger prediction correlation and smaller errors. For BPANN model, R and root-mean square error (RMSE) achieved 0999 as well as 0016 mg in training and 0912 as well as 0207 mg. Therefore, the combination of THz technique and statistical methods represents high precision of prediction and really can act as an effective tool to monitor PM25 together with other normal approaches.
姜晨, 詹洪磊, 李倩, 赵昆, 张振伟, 张存林. 基于太赫兹光谱-统计分析的大气PM2.5监测研究[J]. 光谱学与光谱分析, 2018, 38(2): 643. JIANG Chen, ZHAN Hong-lei, LI Qian, ZHAO Kun, ZHANG Zhen-wei, ZHANG Cun-lin. Terahertz-Statistics-Dependent Approach for Monitoring PM2.5 in Air[J]. Spectroscopy and Spectral Analysis, 2018, 38(2): 643.