留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils

HUANG Yao ZHAO Nan-jing MENG De-shuo ZUO Zhao-lu CHENG Zhao CHEN Yu-nan CHEN Xiao-wei GU Yan-hong

黄尧, 赵南京, 孟德硕, 左兆陆, 程钊, 陈宇男, 陈晓伟, 谷艳红. 土壤多组分PAHs激光诱导二维荧光光谱定量方法研究[J]. 中国光学. doi: 10.37188/CO.2020-0059
引用本文: 黄尧, 赵南京, 孟德硕, 左兆陆, 程钊, 陈宇男, 陈晓伟, 谷艳红. 土壤多组分PAHs激光诱导二维荧光光谱定量方法研究[J]. 中国光学. doi: 10.37188/CO.2020-0059
HUANG Yao, ZHAO Nan-jing, MENG De-shuo, ZUO Zhao-lu, CHENG Zhao, CHEN Yu-nan, CHEN Xiao-wei, GU Yan-hong. Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils[J]. Chinese Optics. doi: 10.37188/CO.2020-0059
Citation: HUANG Yao, ZHAO Nan-jing, MENG De-shuo, ZUO Zhao-lu, CHENG Zhao, CHEN Yu-nan, CHEN Xiao-wei, GU Yan-hong. Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils[J]. Chinese Optics. doi: 10.37188/CO.2020-0059

土壤多组分PAHs激光诱导二维荧光光谱定量方法研究

doi: 10.37188/CO.2020-0059
详细信息
  • 中图分类号: O657.319

Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils

Funds: Supported by National Natural Science Foundation of China (61705238), Anhui Key Research and Development Program(904038316006)
More Information
    Author Bio:

    HUANG Yao (1991—), Male, born in Xinyang City of Henan province, Ph.D Candidate, University of Science and Technology of China. He got his master's degree from Nanchang University in 2017. His research interests are on spectral detection and analysis of pollutants in soils. E-mail: yhuang@aiofm.ac.cn

    ZHAO Nanjing (1976—), Male, born in Dangshan County of Anhui province, PhD, professor. He got his PhD from Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences in 2005. His research interests are on new methods and techniques for environmental optics. E-mail: njzhao@aiofm.ac.cn

    Corresponding author: njzhao@aiofm.ac.cn
  • 摘要: 激光诱导荧光技术具有实时、快速的优势,并且无需样品预处理,是土壤多环芳烃定量分析检测的一种重要分析手段。然而土壤中多环芳烃种类繁多,激光诱导荧光光谱重叠严重,在无法化学分离的情况下实现基体效应复杂土壤中单一多环芳烃的定量是难点之一。本文采用266 nm可移动激光诱导荧光系统获取了农田土壤多环芳烃的荧光光谱,研究了基于单变量线性回归、加权非负最小二乘多元线性回归和支持向量回归的多组分多环芳烃定量分析方法。结果表明:采用单变量线性回归,蒽和菲的相关系数均小于0.90,平均相对误差均大于20%;相比于单变量线性回归,加权非负最小二乘多元线性回归提高了两组分多环芳烃污染土壤中蒽和菲的预测精度,但在多组分多环芳烃污染土壤中平均相对误差仍在20%以上。最后采用GWO-DE优化的支持向量机回归模型分析了多组分多环芳烃污染土壤中的蒽和菲,蒽的平均相对误差由多元线性回归的23.1%下降至5.02%,菲的平均相对误差从20.8%下降到4.83%。该研究为提高土壤多组分多环芳烃激光诱导荧光定量分析的准确性提供了方法支撑。
  • Figure  1.  Schematic diagram of the LIF experimental setup

    Figure  2.  Photograph of the mobile LIF system

    Figure  3.  LIF spectra of multicomponent PAHs in soils

    Figure  4.  The calibration curve of PAHs by MLR

    Figure  5.  The optimized 3D view for AN by grid search

    Figure  6.  Optimization process of GWO-DE model for AN

    Figure  7.  The calibration curve of PAHs by SVR model

    Table  1.   Characteristic fluorescence peaks for univariate regression

    PAHsanalytecharacteristic fluorescence peak
    AN+PYAN404 nm
    PY484 nm
    PHE+PYPHE347 nm
    PY484 nm
    AN+PY+PHEAN404 nm
    PY484 nm
    PHE347 nm
    下载: 导出CSV

    Table  2.   Results of univariate quantitative analysis

    PAHsanalytecorrelation coefficientaverage relative error (%)
    AN+PYAN0.86723.5
    PY0.95614.9
    PHE+PYPHE0.88224.7
    PY0.96113.0
    AN+PY+PHEAN0.89625.1
    PY0.93916.1
    PHE0.87326.4
    下载: 导出CSV

    Table  3.   Characteristic fluorescence peaks or data points for multivariate regression

    PAHsanalytecharacteristic fluorescence peaks(nm)
    AN+PYAN384, 393, 396, 400,404, 406, 408, 416
    PY476, 480, 484, 488, 492
    PHE+PYPHE345, 347, 349, 361, 365, 372, 381, 385
    PY476, 480, 484, 488, 492
    AN+PY+PHEAN384, 385, 393, 396, 400, 404, 406, 408
    PY476, 480, 484, 488, 492
    PHE347, 363, 365, 367, 380, 384, 385, 390
    下载: 导出CSV

    Table  4.   Results of MLR quantitative analysis

    PAHsanalytecorrelation coefficientaverage relative error (%)
    AN+PYAN0.97511.7
    PY0.97712.4
    PHE+PYPHE0.96711.8
    PY0.96312.4
    AN+PY+PHEAN0.91023.1
    PY0.95215.9
    PHE0.93820.8
    下载: 导出CSV
  • [1] ABDEL-SHAFY H I, MANSOUR M S M. A review on polycyclic aromatic hydrocarbons: source, environmental impact, effect on human health and remediation[J]. Egyptian Journal of Petroleum, 2016, 25(1): 107-123. doi:  10.1016/j.ejpe.2015.03.011
    [2] KIM K H, JAHAN S A, KABIR E, et al. A review of airborne polycyclic aromatic hydrocarbons (PAHs) and their human health effects[J]. Environment International, 2013, 60: 71-80. doi:  10.1016/j.envint.2013.07.019
    [3] ABBAS I, BADRAN G, VERDIN A, et al. Polycyclic aromatic hydrocarbon derivatives in airborne particulate matter: sources, analysis and toxicity[J]. Environmental Chemistry Letters, 2018, 16(2): 439-475. doi:  10.1007/s10311-017-0697-0
    [4] ODABASI M, OZGUNERGE FALAY E, TUNA G, et al. Biomonitoring the spatial and historical variations of persistent organic pollutants (POPs) in an industrial region[J]. Environmental Science &Technology, 2015, 49(4): 2105-2114.
    [5] BI X, LUO W, GAO J J, et al. Polycyclic aromatic hydrocarbons in soils from the Central-Himalaya region: distribution, sources, and risks to humans and wildlife[J]. Science of the Total Environment, 2016, 556: 12-22. doi:  10.1016/j.scitotenv.2016.03.006
    [6] SUMAN S, SINHA A, TARAFDAR A. Polycyclic aromatic hydrocarbons (PAHs) concentration levels, pattern, source identification and soil toxicity assessment in urban traffic soil of Dhanbad, India[J]. Science of the Total Environment, 2016, 545-546: 353-360. doi:  10.1016/j.scitotenv.2015.12.061
    [7] HUMEL S, SCHMIDT S N, SUMETZBERGER-HASINGER M, et al. Enhanced accessibility of polycyclic aromatic hydrocarbons (PAHs) and heterocyclic PAHs in industrially contaminated soil after passive dosing of a competitive sorbate[J]. Environmental Science &Technology, 2017, 51(14): 8017-8026.
    [8] KUŚMIERZ M, OLESZCZUK P, KRASKA P, et al. Persistence of polycyclic aromatic hydrocarbons (PAHs) in biochar-amended soil[J]. Chemosphere, 2016, 146: 272-279. doi:  10.1016/j.chemosphere.2015.12.010
    [9] JONES K C, STRATFORD J A, TIDRIDGE P, et al. Polynuclear aromatic hydrocarbons in an agricultural soil: long-term changes in profile distribution[J]. Environmental Pollution, 1989, 56(4): 337-351. doi:  10.1016/0269-7491(89)90079-1
    [10] SHANG Q B, QUAN Y H, XU L SH, et al. Spatial distribution and genesis of polycyclic aromatic hydrocarbons (PAHs) in the surface soil in China[J]. Journal of Ecology and Rural Environment, 2019, 35(7): 917-924.
    [11] LI Y, CUI X, LI ZH CH, et al. Purification and enrichment of polycyclic aromatic hydrocarbons pollutions in oil-field water by column clean-up coupled with dispersive liquid-liquid microextraction[J]. Chinese Journal of Analytical Chemistry, 2018, 46(5): 787-795.
    [12] WANG CH H, WU SH H, ZHOU SH L, et al. Characteristics and source identification of polycyclic aromatic hydrocarbons (PAHs) in urban soils: a review[J]. Pedosphere, 2017, 27(1): 17-26. doi:  10.1016/S1002-0160(17)60293-5
    [13] LI X L, LI Q X, AN SH Q, et al. Determination of fluorine in soil sample by X-ray fluorescence spectrometry[J]. Chinese Journal of Analytical Chemistry, 2019, 47(11): 1864-1869.
    [14] KANG Q W, ZHANG G, CHAI R T, et al. Synthesis of carbon nanodots for detection of aluminum ion with fluorescence enhancement[J]. Chinese Journal of Analytical Chemistry, 2019, 47(12): 1901-1908.
    [15] LEE C K, KO E J, KIM K W, et al. Partial least square regression method for the detection of polycyclic aromatic hydrocarbons in the soil environment using laser-induced fluorescence spectroscopy[J]. Water,Air,and Soil Pollution, 2004, 158(1): 261-275. doi:  10.1023/B:WATE.0000044858.39836.e2
    [16] OKPARANMA R N, MOUAZEN A M. Determination of total petroleum hydrocarbon (TPH) and polycyclic aromatic hydrocarbon (PAH) in soils: a review of spectroscopic and nonspectroscopic techniques[J]. Applied Spectroscopy Reviews, 2013, 48(6): 458-486. doi:  10.1080/05704928.2012.736048
    [17] YANG R J, SHANG L P, BAO ZH B, et al. Feasibility of using laser-induced fluorescence to detect directly polycyclic aromatic hydrocarbons in soil[J]. Spectroscopy and Spectral Analysis, 2011, 31(8): 2148-2150.
    [18] HE J, SHANG L P, DENG H, et al. Detection system of 250 nm laser induce fluorescence of polycyclic aromatic hydrocarbons in soil[J]. Opto-Electronic Engineering, 2014, 41(9): 51-55.
    [19] SARKAR S K. Some multivariate linear regression testing problems with additional observations[J]. Journal of Multivariate Analysis, 1981, 11(4): 556-567. doi:  10.1016/0047-259X(81)90096-8
    [20] BASAK D, PAL S, PATRANABIS D C. Support vector regression[J]. Neural Information Processing-Letters and Reviews, 2007, 11(10): 203-224.
    [21] LU H X, XU M CH, ZHANG W D, et al. Identification of citrus Huanglongbing based on contractive auto-encoder combined extreme learning machine[J]. Chinese Journal of Analytical Chemistry, 2019, 47(5): 652-660.
    [22] FENG Y, PENG CH Y, ZHANG SH W, et al. Preparation of graphene oxide grafted silica monolith and application in determination of polycyclic aromatic hydrocarbons[J]. Chinese Journal of Analytical Chemistry, 2019, 47(6): 814-822.
    [23] ZHAI M Y, ZHAO Y, GAO H, et al. Quantitative study on articular cartilage by fourier transform infrared spectroscopic imaging and support vector machine[J]. Chinese Journal of Analytical Chemistry, 2018, 46(6): 896-901.
    [24] MIRJALILI S, MIRJALILI S M, LEWIS A. Grey wolf optimizer[J]. Advances in Engineering Software, 2014, 69: 46-61. doi:  10.1016/j.advengsoft.2013.12.007
    [25] ZHU A J, XU CH P, LI ZH, et al. Hybridizing grey wolf optimization with differential evolution for global optimization and test scheduling for 3D stacked SoC[J]. Journal of Systems Engineering and Electronics, 2015, 26(2): 317-328. doi:  10.1109/JSEE.2015.00037
  • [1] DONG Lin-xiu, CHEN Zhi-hui, YANG Yi-biao, FEI Hong-ming, LIU Xin.  Far-field range fluorescence enhancement by a hybrid metal-dielectric structure . 中国光学, doi: 10.3788/CO.20201302.0372
    [2] YANG Qiu-jie, HE Zhi-ping, MI Zhong-liang.  Diffraction characteristics analysis of multi-depth phase modulation grating in terahertz band . 中国光学, doi: 10.3788/CO.2019-0147
    [3] YU Chun-feng, CHEN Zhi-chao, JIA Ping, WANG Nai-xiang, HOU Han.  Analysis of the effect of lens shutter on image motion in aerial camera . 中国光学, doi: 10.3788/CO.2019-0127
    [4] GU Yan-hong, ZUO Zhao-lu, ZHANG Zhen-zhen, SHI Chao-yi, GAO Xian-he, LU Jun.  Algorithmic study of total petroleum hydrocarbons in contaminated soil by three-dimensional excitation-emission matrix fluorescence spectroscopy . 中国光学, doi: 10.37188/CO.2019-0216
    [5] Rui TANG, Zi-ye GAO, Zheng-mao WU, Guang-qiong XIA.  Output characteristics of diode-pumped passively Q-switched Yb: CaYAlO4 pulsed laser based on a SESAM . 中国光学, doi: 10.3788/CO.20191201.0167
    [6] Jun-kai SHI, Guo-ming WANG, Rong-yi JI, Wei-hu ZHOU.  Compact dual-wavelength continuous-wave Er-doped fiber laser . 中国光学, doi: 10.3788/CO.20191204.0810
    [7] Yu-qian LIU, He ZHANG, Liang JIN, Ying-tian XU, Hai-zhu WANG, Yong-gang ZOU, Xiao-hui MA, Yan LI.  Compact quasi continuous pumped Nd: YAG Q-switched solid laser . 中国光学, doi: 10.3788/CO.20191202.0413
    [8] Zhi YU, Wei-li YU, Chun-lei GUO.  Charge transfer induced surface enhanced Raman scattering of single crystal and polycrystal perovskites . 中国光学, doi: 10.3788/CO.20191205.0952
    [9] Hua-ming ZHANG, Wei-tao LÜ, Yang ZHANG, Yi-jun ZHANG, Yan-feng FAN, Qi QI.  Analysis of radiation evolution characteristics of the artificial triggered lightning channel . 中国光学, doi: 10.3788/CO.20191203.0670
    [10] Cheng XIAO, Zhi-bin CHEN, Meng-ze QIN, Dong-xiao ZHANG.  SERS characteristics analysis of composite Ag/SiO2 sinusoidal grating . 中国光学, doi: 10.3788/CO.20191201.0059
    [11] Min ZHANG, Jin-guang LV, Jing-qiu LIANG, Zhong-zhu LIANG, Yu-xin QIN, Wei-biao WANG.  Error analysis and fabrication of low-stepped mirrors . 中国光学, doi: 10.3788/CO.20191204.0791
    [12] Chang-bin ZHENG, Jun-feng SHAO, Xue-lei LI, Hua-long WANG, Chun-rui WANG, Fei CHEN, Ting-feng WANG, Jin GUO.  Femtosecond pulsed laser induced damage characteristics on Si-based multi-layer film . 中国光学, doi: 10.3788/CO.20191202.0371
    [13] 于海洋, 涂浪平, 张友林, 赵慧颖, 孔祥贵.  溶剂中稀土上转换纳米粒子表面猝灭效应的定量分析 . 中国光学, doi: 10.3788/CO.20191206.1288
    [14] Ya-ru HAO, Zhao-qi DENG.  High directional multi-beam organic laser . 中国光学, doi: 10.3788/CO.20181104.0576
    [15] Tian-qi LI, Xiao-jie MAO, Jian LEI, Guo-jiang BI, Dong-sheng JIANG.  Analysis and comparison of solid-state lasers and fiber lasers on the coupling of rod-type photonic crystal fiber . 中国光学, doi: 10.3788/CO.20181106.0958
    [16] Hong-yan LIU, Qiang SUN.  Finite element analysis of infrared thermal imaging for four-layers structure of human thigh . 中国光学, doi: 10.3788/CO.20181102.0237
    [17] Chen LI, Razvan STOIAN, Guang-hua CHENG.  Laser-induced periodic surface structures with ultrashort laser pulse . 中国光学, doi: 10.3788/CO.20181101.0001
    [18] 李安, 王亮伟, 郭帅, 刘瑞斌.  激光诱导击穿光谱增强机制及技术研究进展 . 中国光学, doi: 10.3788/CO.20171005.0619
  • 加载中
图(7) / 表 (4)
计量
  • 文章访问数:  14
  • HTML全文浏览量:  1
  • PDF下载量:  0
  • 被引次数: 0
出版历程

Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils

doi: 10.37188/CO.2020-0059
    通讯作者: njzhao@aiofm.ac.cn
  • 中图分类号: O657.319

摘要: 激光诱导荧光技术具有实时、快速的优势,并且无需样品预处理,是土壤多环芳烃定量分析检测的一种重要分析手段。然而土壤中多环芳烃种类繁多,激光诱导荧光光谱重叠严重,在无法化学分离的情况下实现基体效应复杂土壤中单一多环芳烃的定量是难点之一。本文采用266 nm可移动激光诱导荧光系统获取了农田土壤多环芳烃的荧光光谱,研究了基于单变量线性回归、加权非负最小二乘多元线性回归和支持向量回归的多组分多环芳烃定量分析方法。结果表明:采用单变量线性回归,蒽和菲的相关系数均小于0.90,平均相对误差均大于20%;相比于单变量线性回归,加权非负最小二乘多元线性回归提高了两组分多环芳烃污染土壤中蒽和菲的预测精度,但在多组分多环芳烃污染土壤中平均相对误差仍在20%以上。最后采用GWO-DE优化的支持向量机回归模型分析了多组分多环芳烃污染土壤中的蒽和菲,蒽的平均相对误差由多元线性回归的23.1%下降至5.02%,菲的平均相对误差从20.8%下降到4.83%。该研究为提高土壤多组分多环芳烃激光诱导荧光定量分析的准确性提供了方法支撑。

English Abstract

黄尧, 赵南京, 孟德硕, 左兆陆, 程钊, 陈宇男, 陈晓伟, 谷艳红. 土壤多组分PAHs激光诱导二维荧光光谱定量方法研究[J]. 中国光学. doi: 10.37188/CO.2020-0059
引用本文: 黄尧, 赵南京, 孟德硕, 左兆陆, 程钊, 陈宇男, 陈晓伟, 谷艳红. 土壤多组分PAHs激光诱导二维荧光光谱定量方法研究[J]. 中国光学. doi: 10.37188/CO.2020-0059
HUANG Yao, ZHAO Nan-jing, MENG De-shuo, ZUO Zhao-lu, CHENG Zhao, CHEN Yu-nan, CHEN Xiao-wei, GU Yan-hong. Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils[J]. Chinese Optics. doi: 10.37188/CO.2020-0059
Citation: HUANG Yao, ZHAO Nan-jing, MENG De-shuo, ZUO Zhao-lu, CHENG Zhao, CHEN Yu-nan, CHEN Xiao-wei, GU Yan-hong. Study on quantitative methods of laser-induced two-dimensional fluorescence spectroscopy of multicomponent PAHs in soils[J]. Chinese Optics. doi: 10.37188/CO.2020-0059
    • Polycyclic aromatic hydrocarbons(PAHs)are a group of chemicals containing two or more single or fused aromatic rings, which are mainly generated during the incomplete combustion of fuels and organic materials. PAHs can reach soil systems through dry or wet deposition processes. Some of these PAHs are from nearby sources, such as the exhaust gases of cars from adjacent roadways. Other PAHs are originated from more distant sources and have been carried through the air[1]. PAHs can cause diseases, including cancer or transgenation[2-3]. Consequently, more than ten of them are listed as pollutants under priority monitoring, and their concentrations are controlled by many countries and regions[4-6]. PAHs in the atmosphere and water could enter into the soil[7-8], and pollution with PAHs could become a serious environmental issue of the soils. Jones et al.[9] analyzed soil profile samples collected from the same plot and found that the total PAHs burden of the plough layer (0–23 cm) had increased approximately four-to five-fold since the 1890s. Shang et al.[10] found that soils in China moderately polluted, slightly polluted, and uncontaminated with PAHs accounted for approximately 22.6%, 71.1% and 6.3%, respectively.

      Traditional analytical methods for the quantitative analysis of PAHs in soils usually need pretreatment and chromatographic separation[11]. These analytical methods have the advantages of low detection limits and high sensitivity. However, the pretreatment of samples(including extraction, concentration, and purification) is complicated and costly, some steps being possibly harmful to the operator’s health[12]. Fluorescence spectroscopy has become a powerful analytical technique for organic contaminants for their high sensitivity, rapid and multicomponent analysis without sample preparation[13-14]. Among them, laser-induced fluorescence(LIF) has already shown excellent potential for in-situ analysis, which can offer real-time and non-destructive measurement, using strong fluorescent signals emitted by the excited orbital electrons of aromatic hydrocarbons to identify and quantify PAHs.

      Some researchers have focused on the research of PAHs detection in soils by LIF. Lee et al.[15] employed partial least square regression(PLSR) method and multivariate linear regression(MLR) to analyze normalized data obtained by diffuse reflectance. The LIF concentration data were compared with those measured by high-performance liquid chromatography (HPLC), with values of correlation coefficient (R) of 0.96 and 0.90 for phenanthrene and pyrene in artificially contaminated soil samples, respectively. Aldstadlt et al.[16] reported a sufficiently close match between the predicted and measured PAHs by using PLSR analysis in Rapid Optical Screening Tool laser-induced fluorescence(ROST LIF) system, as this system was designed for qualitative applications. Yang et al.[17] used a 355 nm Nd:YAG laser and fiber optic spectrometer of thermoelectric refrigeration to detect anthracene in soils and established a linear relationship between the concentration of anthracene and the intensity of fluorescence. Furthermore, He et al.[18] established a laser-induced detection, based on 250 nm laser to improve the detection limit and the linear relationship of PAHs in soils.

      The matrixes of the measured soil samples are complex, and fluorescence spectra of multicomponent PAHs could often overlap. Laser-induced fluorescence spectra of PAHs are two-dimensional spectral data, and the algorithm of analysis for such data is called two-dimensional correction or first-order correction algorithm. MLR is a classic two-dimensional correction algorithm[19]. It can realize the quantitative analysis of multiple components in the system without complete separations under defined conditions. However, the components which could interfere in samples are unknown and complicated, which may cause poor linearity of fluorescence intensity and concentration in the sample. Therefore, it is still difficult to accurately predict the concentration of a specific substance in complicated multicomponent systems. Support Vector Regression(SVR)[20-21] algorithm is a machine learning algorithm, based on statistical learning theory, which is particularly suitable for nonlinear quantitative analysis of small sample data.

      To the best knowledge of us, quantitative analysis of multicomponent PAHs in the soil with complicated background has not been reported. Thus, in the present paper, the potential of MLR and SVR as calibration methods for multicomponent PAHs determination in soils, by a mobile LIF system without separation was evaluated. The agricultural soil samples were collected in the field as the calibration set to test the accuracy of the MLR and SVR model. For this purpose, the concentrations of PAHs in a series of soil samples were analyzed.

    • The schematic diagram of the LIF setup is presented in Fig.1. A Q-switched Nd:YAG laser(Brilliant Quantel)working at 266 nm was used as the excitation light source, with a repetition rate of 10 Hz. The laser was reflected by the 266 nm reflector mirror and expanded by the beam expander. The pulse energy was about 80 mJ, and the corresponding laser energy density for the LIF measurement was 0.88 mJ/cm2. The soil sample was rotated by a rotary platform slowly to avoid the focusing of the laser on the same area. The fluorescence emission was coupled into an optical fiber and the LIF spectra were detected by a spectrometer (Ocean Maya2000pro), which enabled us to detect the wavelength ranging from 200 nm to 1100 nm with a spectral resolution of 0.5 nm. All of the instruments were assembled on a mobile platform that could be used in the laboratory or field (Fig.2).

      Figure 1.  Schematic diagram of the LIF experimental setup

      Figure 2.  Photograph of the mobile LIF system

      Agricultural soil samples were collected from Tongling city, Anhui Province, China, by our research group. All the soil samples were grounded, and sieved through 100-mesh to remove roots. Anthracene(AN), pyrene(PY) and phenanthrene(PHE) are among the 16 PAHs under the priority control of the US Environmental Protection Agency[22]. They were weighed, dissolved, and then added into agricultural soils to prepare artificially contaminated samples with multicomponent PAHs. For all samples, 10 spectra were accumulated into a single analytical spectrum to improve the signal-to-noise ratio, and every sample was measured 10 times under the same conditions to reduce measurement errors.

    • Fig.3(a) shows the LIF spectrum of soil contaminated with PY and PHE. The fluorescence characteristic peaks of PHE near 365 nm and 385 nm were significantly affected by the main peaks of PY, and the resolution of the spectral peak was poor. Fig.3(b) shows the laser-induced fluorescence spectrum of soil contaminated with AN and PY. Three characteristic fluorescence peaks of AN(near 384 nm, 404 nm and 428 nm) were all disturbed by the fluorescence peaks of PY, which was difficult to distinguish.

      Figure 3.  LIF spectra of multicomponent PAHs in soils

      The fluorescence characteristic peaks used for univariate analysis should avoid the interference of other fluorescence peaks. Characteristic fluorescence peak spectral lines are shown in Tab.1, and results of univariate quantitative analysis are shown in Tab.2.

      Table 1.  Characteristic fluorescence peaks for univariate regression

      PAHsanalytecharacteristic fluorescence peak
      AN+PYAN404 nm
      PY484 nm
      PHE+PYPHE347 nm
      PY484 nm
      AN+PY+PHEAN404 nm
      PY484 nm
      PHE347 nm

      Table 2.  Results of univariate quantitative analysis

      PAHsanalytecorrelation coefficientaverage relative error (%)
      AN+PYAN0.86723.5
      PY0.95614.9
      PHE+PYPHE0.88224.7
      PY0.96113.0
      AN+PY+PHEAN0.89625.1
      PY0.93916.1
      PHE0.87326.4

      In the soil contaminated by AN and PY, the correlation coefficient of AN was 0.867, and the average relative error was 23.5%, while in the case of contamination by PHE and PY, the correlation coefficient of PHE was 0.882, and the average relative error was 24.7%. The correlation coefficients for AN and PHE were 0.896 and 0.873, while the average errors were 25.1% and 26.4% in the soil contaminated with multicomponent PAHs.

    • The LIF spectra of PAHs could be generally regarded as the linear superposition of the fluorescence spectra of multiple components. Since the concentration of PAHs in soils cannot be negative values, weighted non-negative least square was introduced for multivariate linear regression, which takes into account the non-negative concentration of PAHs and the dispersion of spectra in different wavelengths. In order to reduce the amount and to improve the efficiency of calculations, n points near the characteristic peak of the fluorescence spectra were selected to characterize the fluorescence spectra of PAHs. Therefore, the fluorescence spectra of possible PAHs in soils can be expressed as:

      $$ {e_k} = ({e_{k1}},{e_{k2}},\cdots {e_{kn}}) $$ (1)

      A LIF spectrum of the sample can also be expressed as:

      $$ E = ({E_1},{E_2},\cdots {E_n}) $$ (2)

      The contribution of each possible substance to the fluorescence spectra can be expressed as ak, and Equation 2 can be expressed as

      $$ E = \sum {{a_k}} * {e_k} $$ (3)

      There may be a linear superposition of the fluorescence spectrum of the substance. However, this linear superposition would be affected by soil properties. Therefore, a weight factor ${w_{ki}}$ was introduced when using the least square algorithm. The total error of this calculation method can be expressed as:

      $$ {y^2}(a,a,\cdots {a_k}) = \sum\nolimits_{i = 1}^{{n_i}} {\left( {\frac{{{E_i} - \sum\nolimits_{k = 1}^{{n_k}} {{a_k} * {e_{ki}}} }}{{\sum\nolimits_{k = 1}^{{n_k}} {{a_k} * {w_{ki}}} }}} \right)} $$ (4)

      The total error is a nonlinear function, which could be solved iteratively.

      Tab.3 lists the characteristic fluorescence peaks or data points selected for multivariate regression analysis of multicomponent PAHs in soils.

      Tab.4 shows the results of MLR quantitative analysis. As shown in Fig.4(a), the correlation coefficient of PHE in soils contaminated with PHE and PY was 0.967, while the average relative error was 11.8%. As shown in Fig.4(b), the correlation coefficient of AN in the soil contaminated with AN and PY was 0.975, while the average relative error was 11.7%. The results showed that the calibration accuracy for soil samples could be improved by considering the influence of appropriate multivariate lines in the quantitative analysis. It worth noting that even with multivariate linear regression, the average relatives of AN and PHE in soils contaminated with multicomponent PAHs were still over 20%. Due to the influence of soil matrix and self-absorption effects, there may be a nonlinear relationship between concentrations and the intensities of LIF, especially in the soil contaminated with multicomponent PAHs. Neither univariate linear regression, nor MLR can solve the nonlinear relationship. Therefore, a precise calibration model was assigned to establish the efficient quantification of the concentration of PAHs in unknown samples.

      Table 3.  Characteristic fluorescence peaks or data points for multivariate regression

      PAHsanalytecharacteristic fluorescence peaks(nm)
      AN+PYAN384, 393, 396, 400,404, 406, 408, 416
      PY476, 480, 484, 488, 492
      PHE+PYPHE345, 347, 349, 361, 365, 372, 381, 385
      PY476, 480, 484, 488, 492
      AN+PY+PHEAN384, 385, 393, 396, 400, 404, 406, 408
      PY476, 480, 484, 488, 492
      PHE347, 363, 365, 367, 380, 384, 385, 390

      Table 4.  Results of MLR quantitative analysis

      PAHsanalytecorrelation coefficientaverage relative error (%)
      AN+PYAN0.97511.7
      PY0.97712.4
      PHE+PYPHE0.96711.8
      PY0.96312.4
      AN+PY+PHEAN0.91023.1
      PY0.95215.9
      PHE0.93820.8

      Figure 4.  The calibration curve of PAHs by MLR

    • SVR is a machine learning algorithm based on statistical learning theory and has a better ability of regression analysis by finding the best learning ability under limited sample information and structural risk minimization[23]. The samples in low dimensional space are nonlinearly mapped to high dimensional space by using the kernel function expansion method, and the nonlinear regression problem in low dimensional space is solved by the linear method in the new space. Therefore, the SVR model was fit for dealing with small samples and nonlinear LIF quantitative analysis, particularly. A SVR model was established to measure the concentration of AN and PHE in the soil contaminated with multicomponent PAHs. The expression of the SVR model could be described as follows:

      $$ {c_s} = f\left( {{I_1},{I_2},{I_3},\cdots ,{I_n}} \right) $$ (5)

      where $\left( {{I_1},{I_2},{I_3},\cdots ,{I_n}} \right)$ is the measured spectral intensity. The regression function of SVR for spectral analysis could be expressed as:

      $$ {C_s} = f(I) = \sum\limits_{i \in (SV)} {{\alpha _i}} K({I_i},{I_s}) + b $$ (6)

      where $(SV)$ is the input spectral data set, ${\alpha _i}$ is the Lagrange multiplier, $b$ is a constant. ${I_s}$ is the intensity of the characteristic fluorescence peaks or data points of PAHs, which can use the same data as the MLR, while $K$ is the kernel function. The most commonly used kernel function is the radial basis kernel function(RBF), and can be expressed as:

      $$ K({I_i},{I_s}) = {e^{\left( { - \frac{{{{\left| {{I_i} - {I_s}} \right|}^2}}}{{2{\sigma ^2}}}} \right)}} $$ (7)

      where $\dfrac{1}{{2{\sigma ^2}}}$ is an adjustable attributed parameter, also can be expressed as g.

      In a SVR model, two parameters should be adjusted, including the penalty coefficient c and the kernel function parameter g. These two parameters have a great influence on the accuracy, stability and generalization performance of the model, so the optimization of parameters is crucial to a high-performance model.

      There are mainly three optimization pathways for SVR: (1) c and g are selected by experience. This method is rarely used because it depends on users and samples; (2) Grid search optimization; (3) Swarm Intelligent Algorithm. The basic principle of the grid search method is to let c and g traverse all points in the grid within a certain value range. When the interval is large, and the step distance is small, a global optimum can be searched. However, traversing the whole grid would cause a large amount of calculation and the c obtained by this method is large, which easily leads to overfitting and poor generalization of the model. When the interval is small, and the step distance is large, it is easy to miss the optimal global solution. On the other hand, swarm intelligence optimization is fast and random, so it is easy to miss the optimal global solution when c and g are not in a limited range. Therefore, grid search combined with intelligent swarm optimization was used for optimizing parameters in this paper. Gray Wolf Optimization (GWO) is an efficient swarm intelligence algorithm, which was proposed by Mirjalili in 2014[24]. To reduce the possibility of GWO falling into a local optimum, Zhu et al.[25] proposed a hybrid optimizer method combining GWO with differential evolution (DE). In this work, GWO-DE was used for finding the optimal solution precisely.

      Grid search method was used to search the optimal parameters on a large scale roughly. The parameters were set as follows: ranges of penalty parameter c and RBF kernel function parameter g both were 2-10-210, step distance of c and g was 1. Taking AN for example, as shown in Fig.5, the optimal value of c was 45.2548, g was 0.0625, and the mean square error was 3.7458. According to the results found by grid search, parameters of GWO-DE were set as follows: population size n=20, the maximum number of iterations N=30, value range of c and g both were 10-2-102. The crossover probability was 0.2, and upper and lower bounds of the scaling factors of differential evolution were 0.2 and 0.8, respectively.

      The optimization process of parameters c and g by GWO-DE are shown in Fig.6. It can be observed that the GWO-DE model could rapidly converge to the optimal solution after 5 iterations, and the fitness is stable after 15 iterations. The optimal value of c was 17.7071, while in case of g was 0.0183. Then the SVR model with optimized parameters was built, and it was used to analyze the concentration of AN in soils quantitatively. Fig.7(b) shows the prediction results of AN in the SVR model. The same approach was used to build the SVR model for PHE in soils, and the prediction results are shown in Fig.7(a). The correlation coefficients of AN and PHE were 0.995 and 0.998, respectively. Compared with the values of 0.910 and 0.938 obtained by MLR, the results obtained from the SVR model were closer to 1. The average relatives of AN and PHE in the soil contaminated with multicomponent PAHs were over 20% in MLR, while those of AN and PHE were 5.02% and 4.83% in the SVR model, respectively. The results demonstrated that the measurement accuracy of AN and PHE in the soil contaminated with multicomponent PAHs could be efficiently improved by using the SVR model.

      Figure 5.  The optimized 3D view for AN by grid search

      Figure 6.  Optimization process of GWO-DE model for AN

      Figure 7.  The calibration curve of PAHs by SVR model

    • Three different quantitative methods for LIF spectra of multicomponent PAHs in agricultural soil were investigated in the present paper. The results showed that the correlation coefficients of AN and PHE were both less than 0.90, and the average relative errors were higher than 20% by using univariate linear regression. Considering heavily overlapped LIF spectra of multicomponent PAHs, MLR based on weighted non-negative least square was chosen for quantitative analysis. This method improved the accuracy of prediction of AN and PHE in the soil contaminated with bi-component PAHs, but the average relative errors were still over 20% in the soil contaminated with multicomponent PAHs, due to the influence of soil matrix and self-absorption effects. Finally, a SVR model optimized by GWO-DE was applied for the concentration measurement of AN and PHE in agricultural soil contaminated with multicomponent PAHs. The correlation coefficients of AN and PHE were 0.995 and 0.998, which were closer to 1. Compared with MLR, the average relative errors were reduced from 23.1% to 5.02% for AN and from 20.8% to 4.83% for PHE, respectively. The results demonstrated that the accuracy of the measurement of AN and PHE in the soil contaminated with multicomponent PAHs could be improved by using the SVR model. These results gained through this experiment make it possible to apply a mobile LIF system to meet the demands of monitoring PAHs in the field.

参考文献 (25)

目录

    /

    返回文章
    返回