水体COD激光诱导击穿光谱快速测量方法研究
|
摘要
水体化学需氧量(COD)是一个重要的水体质量指标,一般用来衡量有机物的污染程度。对COD的检测长期依赖采样后的实验室化学分析方法,目前应用最普遍的是重铬酸钾氧化法与酸性高锰酸钾氧化法。化学分析的方法操作复杂,耗时费力,且引入新的化学药剂,造成二次污染,因此,急需一种能够实现水体COD快速测量的检测技术。在前期研究基础上,对水体COD的激光诱导击穿光谱检测方法进行深入探索,重点是优化模型预测速度,目的是研究激光诱导击穿光谱技术用于对水体COD的快速测量方法。采集了不同COD浓度的99个水体样本,分为训练集和测试集两组,通过重铬酸钾氧化法测定各水样的COD值,作为真实值,利用实验室自建的激光诱导击穿光谱采集系统采集各水样波长在200～1 000 nm的光谱信息,利用偏最小二乘算法建立训练集水样COD的定量化测量模型,然后对测试集光谱数据进行预测,将预测结果与实验室化学方法测定的真实值进行对比,评估预测效果。通过对原始光谱建立的预测模型进行分析,发现在建模过程中,大量的激光诱导击穿光谱数据与COD浓度相关性很差,而这些无用数据参与计算,浪费了计算资源,延长了检测时间,造成系统负载过大,不利于便携式检测设备的开发。重点研究贡献度最大的前几个主成分,通过对COD测量原理和PLS模型载荷分析,找到LIBS光谱中与水样COD浓度相关性最高的主要特征峰,经过分析发现,主要为来源于水中有机物中的C, H, O, N以及水中一些还原性离子元素的特征峰,这些特征峰对COD的模型预测能力贡献最大,而COD的定义正是衡量水体中这些元素的多少,这与该研究分析结论相吻合。为了实现检测速度的提升,提取这些特征峰,对光谱数据进行降维,剔除大量无关或相关性较低的数据,经过多次筛选和降维,最终将原来参与计算的每条光谱的13 622个数据降到28个,大大降低系统的运算量,却依然能够保留不错的预测能力。筛选出的28个特征波长最能反映水体COD浓度,为水体COD便携式的多波段检测设备的研发,实现对COD的快速测量奠定了基础。
Chemical Oxygen Demand(COD) is an important water quality parameter, which is generally used to reflect the degree of organic pollution. The detection of COD has long relied on laboratory chemical analysis methods after sampling. The most commonly used method is potassium dichromate oxidation and acid Potassium Permanganate oxidation. However, the chemical analysis methodsarecomplicated, time-consuming and labor-intensive. Moreover, the introduction of new chemicals by these methods causes secondary pollution. Therefore, there is an urgent need for a detection technique that enables rapid measurement of COD in water. Based on the previous research, the test method of COD by the laser-induced breakdown spectroscopy was explored in this paper. The focus was to optimize the model prediction speed in order to study therapid measurement of COD in waterbythe techniqueof laser-induced breakdown spectroscopy. We collected 99 water samples with different COD concentrations, and divided them into two groups: training set and testing set. The COD concentration of each water sample was measured by potassium dichromate oxidation method. And the spectral information of each water sample at the wavelengths of 200～1 000 nm was collected by the laser-induced breakdown spectroscopy acquisition system built by our laboratory. Partial least squares(PLS) algorithm was used to establish a quantitative measurement model for COD of training set samples and the spectral data of test set were predicted. The predicted results were compared with the real values measured by laboratory chemical methods to evaluate the predicted results. By analyzing the prediction model established by the original spectrum, it was found that a large number of laser-induced breakdown spectral data have poor correlation with COD concentration during the modeling process, and these useless data participated in the calculation, wasting computing resources, dragging the detection time, causing the system load to be too large, which was not conducive to the development of portable detection equipment. We focused on the first few principal components with the largest contribution. By analyzing the principle of COD measurement and the load of PLS model, we found the main characteristic peaks of LIBS spectrum which have the highest correlation with the concentration of COD in water. These characteristic peaks belong to C, H, O, N and some reductive ion elements in water. Most of C, H, O and N come from organic matter in water whose characteristic peaks have the greatest contribution to the prediction ability of COD model. The definition of COD reflects the amount of these elements in the water body, which is consistent with our analysis conclusion. In order to improve the detection speed, we extracted these characteristic peaks, and eliminated a large number of unrelated or low-correlation data. After many times of screening and dimensionality reduction, the original 13 622 data of each spectrum were reduced to 28, which greatly reduced the computational complexity of the system, but still retains good prediction ability. The 28 characteristic wavelengths selected are the best ones to reflect the concentration of COD in water, which lays a foundation for the development of portable multi-band detection equipment for COD in water and the rapid measurement of COD.
Chemical Oxygen Demand(COD) is an important water quality parameter, which is generally used to reflect the degree of organic pollution. The detection of COD has long relied on laboratory chemical analysis methods after sampling. The most commonly used method is potassium dichromate oxidation and acid Potassium Permanganate oxidation. However, the chemical analysis methodsarecomplicated, time-consuming and labor-intensive. Moreover, the introduction of new chemicals by these methods causes secondary pollution. Therefore, there is an urgent need for a detection technique that enables rapid measurement of COD in water. Based on the previous research, the test method of COD by the laser-induced breakdown spectroscopy was explored in this paper. The focus was to optimize the model prediction speed in order to study therapid measurement of COD in waterbythe techniqueof laser-induced breakdown spectroscopy. We collected 99 water samples with different COD concentrations, and divided them into two groups: training set and testing set. The COD concentration of each water sample was measured by potassium dichromate oxidation method. And the spectral information of each water sample at the wavelengths of 200～1 000 nm was collected by the laser-induced breakdown spectroscopy acquisition system built by our laboratory. Partial least squares(PLS) algorithm was used to establish a quantitative measurement model for COD of training set samples and the spectral data of test set were predicted. The predicted results were compared with the real values measured by laboratory chemical methods to evaluate the predicted results. By analyzing the prediction model established by the original spectrum, it was found that a large number of laser-induced breakdown spectral data have poor correlation with COD concentration during the modeling process, and these useless data participated in the calculation, wasting computing resources, dragging the detection time, causing the system load to be too large, which was not conducive to the development of portable detection equipment. We focused on the first few principal components with the largest contribution. By analyzing the principle of COD measurement and the load of PLS model, we found the main characteristic peaks of LIBS spectrum which have the highest correlation with the concentration of COD in water. These characteristic peaks belong to C, H, O, N and some reductive ion elements in water. Most of C, H, O and N come from organic matter in water whose characteristic peaks have the greatest contribution to the prediction ability of COD model. The definition of COD reflects the amount of these elements in the water body, which is consistent with our analysis conclusion. In order to improve the detection speed, we extracted these characteristic peaks, and eliminated a large number of unrelated or low-correlation data. After many times of screening and dimensionality reduction, the original 13 622 data of each spectrum were reduced to 28, which greatly reduced the computational complexity of the system, but still retains good prediction ability. The 28 characteristic wavelengths selected are the best ones to reflect the concentration of COD in water, which lays a foundation for the development of portable multi-band detection equipment for COD in water and the rapid measurement of COD.
引文
[1] Goh C P,Lim P E.Asean Journal on Science and Technology for Development,2017,25(2):383.
[2] ZHANG Ji-hua(张继华).Environmental Science and Technology(环境科学与技术),2004,27(8):37.
[3] YUAN Han-hong,HU Yan-ping,QI Hui-liang(袁汉鸿,胡艳平,戚惠良).Water Purification Technology(净水技术),2005,24(4):67.
[4] LUO Guo-bing(罗国兵).Rock and Mineral Analysis(岩矿测试),2013,32(6):860.
[5] Pan T,Chen Z,Chen J,et al.Analytical Methods,2012,4(4):1046.
[6] Fortes F J,Moros J,Lucena P,et al.Analytical Chemistry,2013,85(2):640.
[7] Zhang D C,Hu Z Q,Su Y B,et al.Optics Express,2018,26(14):18794.
[8] Costa V C,Aquino F W B,Paranhos C M,et al.Polymer Testing,2017,59:390.
[9] Hoehse M,Paul A,Gornushkin I,et al.Analytical & Bioanalytical Chemistry,2012,402(4):1443.
[10] Janzen C,Fleige R,Noll R,et al.Spectrochimica Acta Part B,Atomic Spectroscopy,2005,60(7):993.
[11] Bilge G,Sezer B,Boyaci I H,et al.Spectrochimica Acta Part B,Atomic Spectroscopy,2018,145.
[12] Mousavi S J,Farsani M H,Darbani S M R,et al.Applied Physics B,2016,122(5):1.
[13] YE Song,GU Ya-hui,DU Xiao-fan,et al(叶松,顾亚辉,杜晓凡,等).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2017,37(11):3585.
[2] ZHANG Ji-hua(张继华).Environmental Science and Technology(环境科学与技术),2004,27(8):37.
[3] YUAN Han-hong,HU Yan-ping,QI Hui-liang(袁汉鸿,胡艳平,戚惠良).Water Purification Technology(净水技术),2005,24(4):67.
[4] LUO Guo-bing(罗国兵).Rock and Mineral Analysis(岩矿测试),2013,32(6):860.
[5] Pan T,Chen Z,Chen J,et al.Analytical Methods,2012,4(4):1046.
[6] Fortes F J,Moros J,Lucena P,et al.Analytical Chemistry,2013,85(2):640.
[7] Zhang D C,Hu Z Q,Su Y B,et al.Optics Express,2018,26(14):18794.
[8] Costa V C,Aquino F W B,Paranhos C M,et al.Polymer Testing,2017,59:390.
[9] Hoehse M,Paul A,Gornushkin I,et al.Analytical & Bioanalytical Chemistry,2012,402(4):1443.
[10] Janzen C,Fleige R,Noll R,et al.Spectrochimica Acta Part B,Atomic Spectroscopy,2005,60(7):993.
[11] Bilge G,Sezer B,Boyaci I H,et al.Spectrochimica Acta Part B,Atomic Spectroscopy,2018,145.
[12] Mousavi S J,Farsani M H,Darbani S M R,et al.Applied Physics B,2016,122(5):1.
[13] YE Song,GU Ya-hui,DU Xiao-fan,et al(叶松,顾亚辉,杜晓凡,等).Spectroscopy and Spectral Analysis(光谱学与光谱分析),2017,37(11):3585.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |