面向XRF的竞争性自适应重加权算法和粒子群优化的支持向量机定量分析研究

doi:10.3964/j.issn.1000-0593(2023)12-3742-05

摘要
参考文献
相关文章 (15)

全文: PDF (3719 KB)
输出: BibTeX | EndNote (RIS)

摘要：研究高效、准确、便捷的土壤重金属检测方法对于了解土壤的污染状况以及开展污染防治工作具有重要的意义。由于X射线荧光光谱分析（XRF）技术具备快速、准确、无损检测、样品制备简单等优势，在土壤重元素定量检测获得广泛应用。XRF仪器测试标准样品的荧光光谱并建立校准曲线，通过反演计算得到待测样品的元素含量。由于样品元素间存在基体效应，以及荧光谱特征峰存在叠加干扰，未经优化的校准曲线的线性度较差，这给反演计算来困难。为了解决上述问题，分别利用小波变换、非对称加权惩罚最小二乘法(arPLS)对光谱进行去噪和扣除本底基线，提高校准曲线的决定系数（R²）；运用竞争性自适应重加权算法（CARS），针对不同目标元素优化变量选取；进一步地，基于选取的变量建立粒子群算法（PSO）优化的支持向量机回归（SVR）模型，并通过该模型反演计算各元素含量，提高定量分析的准确度和预测的泛化能力。实验结果显示，经过小波去噪和arPLS本底扣除后的校准曲线的决定系数（R²）有明显提升，Cr、Cu、Zn、As、Pb分别从0.965、0.979、0.971、0.794、0.915提高为0.979、0.987、0.981、0.828、0.953；通过CARS选取的谱线变量的个数大幅度减少，从2 048个通道降低到30个以下，为原来变量个数的1.5%，提高了变量选择的精准性；与偏最小二乘法（PLS）、未优化的SVR模型进行对比，采用CARS变量选择和PSO优化的SVR模型进行含量预测，训练集R²_C与测试集R²_P的决定系数分别在0.99、0.90以上，预测准确性有明显提高。因此，所提出的竞争性自适应重加权算法和PSO优化的SVR定量分析模型对于土壤重金属元素定量分析具有较好的理论指导和应用价值。

关键词：X射线荧光光谱；土壤重金属；竞争性自适应重加权算法；粒子群算法；支持向量机回归模型

Abstract：Research on efficient, accurate and convenient soil heavy metal detection methods is of great significance for understanding soil pollution and carrying out pollution prevention and control. Because X-ray fluorescence spectrometry (XRF) technology has the advantages of being fast, accurate and non-destructive, it has been widely used in detecting element content. The XRF method obtains the concentration of the sample to be tested by measuring the fluorescence intensity of the sample to be tested, and establishing a corresponding relationship using the fluorescence intensity of the standard sample of the calibration curve and the corresponding concentration. However, due to the existence of matrix effect and spectral line overlapping interference, the element spectral line intensity obtained in the actual XRF analysis test and its corresponding concentration do not show a relatively perfect linear relationship. In order to solve the above problems, this paper uses wavelet transform and asymmetric weighted penalized least squares (arPLS) to denoise the spectrum and correct the baseline, which improves the determination coefficient of the calibration curve to a certain extent. The characteristic energy spectral line selection model of different heavy metal elements was constructed by the Competitive Adaptive Reweighting Algorithm (CARS) algorithm to explore the aggregation performance of the characteristic spectral lines. Further, based on the selected features, the particle swarm optimization (PSO) optimized support vector machine regression (SVR) model is used to predict the element content, and the generalization ability of the quantitative analysis model is improved. Partial least squares regression (PLSR) and SVR models are used to compare. The results show that: after pretreatment, the coefficients of determination of the calibration curves of Cr, Cu, Zn, As, Pb are improved from 0.965, 0.979, 0.971, 0.794, 0.915 to 0.979, 0.987, 0.981, 0.828, 0.953; the characteristic lines selected by CARS In addition to the elements to be analyzed, some also correspond to the soil matrix effect elements and the corresponding spectral line interference elements, which shows the effectiveness of the CARS algorithm in feature selection, and the number of variables has changed from 2 048 to 9~29, which is 0.43%~1.42% of the original number of variables, which makes the variables of feature selection more statistical and intelligent; the content prediction using the PSO-optimized SVR model is higher than the accuracy of SVR and PLSR, training set and test set. The coefficients of determination are above 0.99 and 0.89, respectively.

Key words：X-ray fluorescence spectroscopy;Soil heavy metals;Competitive adaptive reweighted sampling;Particle swarm optimization;Support vector regression

收稿日期: 2022-04-23 修订日期: 2022-10-20

中图分类号:

O434.13

基金资助: 国家自然科学基金项目(62075028)资助

通讯作者: 李福生 E-mail: lifusheng@uestc.edu.cn

作者简介: 程惠珠，女，1976年生，电子科技大学自动化工程学院教授 e-mail: 15651006609@163.com

引用本文:

程惠珠，杨婉琪，李福生，马骞，赵彦春. 面向XRF的竞争性自适应重加权算法和粒子群优化的支持向量机定量分析研究[J]. 光谱学与光谱分析, 2023, 43(12): 3742-3746.
CHENG Hui-zhu, YANG Wan-qi, LI Fu-sheng, MA Qian, ZHAO Yan-chun. Genetic Algorithm Optimized BP Neural Network for Quantitative Analysis of Soil Heavy Metals in XRF. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(12): 3742-3746.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2023)12-3742-05 或 https://www.gpxygpfx.com/CN/Y2023/V43/I12/3742

[1] FENG Yong-jie（冯永杰）. Environmental and Development（环境与发展），2020，32(4)：77.
[2] Zhou Y，Aamir M，Liu K，et al. Environmental Pollution，2018，240：116.
[3] Khan S，Naushad M，Lima E C，et al. Journal of Hazardous Materials，2021，417: 126039.
[4] YU Tao，JIANG Tian-yu，LIU Xu，et al（余涛，蒋天宇，刘旭，等）. Geology in China（中国地质），2021，48(2)：460.
[5] AN Qi-qi（安琪琪）. Modern Agricultural Science and Technology（现代农业科技），2020，17：166.
[6] ZHANG Lian-xiang，FU Bin（章连香，符斌）. Chinese Journal of Inorganic Analytical Chemistry（中国无机分析化学），2013，3：1.
[7] Gardner R P，Li F S. X-Ray Spectrometry，2011，40(6)：405.
[8] Li F S, Yang W Q，Ma Q，et al. Measurement Science & Technology，2021，32(10)：105501.
[9] REN Shun，ZHANG Xiong，REN Dong，et al（任顺，张雄，任东，等）. Journal of Instrumental Analysis（分析测试学报），2020，39(7)：829.
[10] CHEN Ying，LIU Zheng-ying，XIAO Chun-yan，et al（陈颖，刘峥莹，肖春艳，等）. Spectroscopy and Spectral Analysis（光谱学与光谱分析），2021，41(7)：2175.
[11] TANG Hai-tao，MENG Xiang-tian，SU Xun-xin，et al（唐海涛，孟祥添，苏循新，等）. Transactions of the Chinese Society of Agricultural Engineering（农业工程学报），2021，37(2)：105.
[12] CHEN Yu，QIU Zhi-jun，ZHANG Bin（陈煜，邱智军，张彬）. Journal of Instrumental Analysis（分析测试学报），2021，40(12)：1004.
[13] LIU Jin，XU Wen-li，SUN Tong，et al（刘津，许文丽，孙通，等）. Chinese Journal of Analysis Laboratory（分析试验室），2018，37(1)：1.
[14] Li Hongdong，Liang Yizeng，Xu Qingsong，et al. Analytica Chimica Acta，2009，648(1)，77.
[15] Da Silva D J，Wiebeck H. Journal of Polymer Research，2018，25(5): 112.
[16] HONG Qian，ZHAO Jin-hui，YUAN Hai-chao，et al（洪茜，赵进辉，袁海超，等）. Chinese Journal of Analysis Laboratory（分析试验室），2013，32(12)：6.
[17] GUO Yang，GUO Jun-xian，SHI-Yong，et al（郭阳，郭俊先，史勇，等）. Food & Machinery（食品与机械），2021，37(6)：81.