多变量回归对大米中Cd元素LIBS分析精度比较

Comparison of precision and accuracy in analyzing Cd in rice by LIBS combined with multivariate regression

为了验证多变量回归分析提高激光诱导击穿光谱(LIBS)技术对大米中重金属元素Cd分析精度的可行性,在实验室条件下,对市售大米进行40个不同浓度梯度的氯化镉(Cd Cl2·52H2O)溶液浸泡,并对Cd污染大米进行干燥粉碎压片处理。采用优化后的LIBS系统参数采集压片大米中Cd元素的光谱信息,再利用阳极溶出伏安法(ASV)获取大米中Cd元素的真实含量。选取变量在211.03~299.96nm波长范围,运用偏最小二乘(PLS)与最小二乘支持向量机(LSSVM)对LIBS光谱信息与Cd真实浓度进行回归分析。PLS与LSSVM两种模型的定标集相关系数分别为0.9936,0.9992,验证集相关系数分别为0.9866,0.9946,交叉验证均方根误差RM SCEV分别为10.53,11.59,预测均方根误差RM SEP分别为10.52,10.80;定标集平均相对误差分别为17.2%,4.2%,验证集平均相对误差分别为12.1%,8.6%。试验结果表明,PLS与LSSVM两种方法均能对大米中Cd元素进行精确预测,且LSSVM的分析稳定性更好。本工作为大米中重金属Cd的LIBS快速、精确检测提供了理论参考。

The aim in this work is to verify the feasibility of multivariate regression in analyzing Cd in rice by observing the spectra obtained via LIBS（ laser induced breakdown spectroscopy）. The rice from local market was dipped into Cd solution with 40 different concentrations by mixing cadmium chloride（ CdCl2·5/2H2O） and deionized water. The polluted rice was dried totally,ground and pressed to pellets to eliminate the influence of water content. The LIBS spectra of samples were collected at the optimized conditions, and the real contents of Cd in rice were determined by anodic stripping voltammetry（ ASV）. The range of wavelength from 211. 03 nm to 299. 96 nm including CdⅠ228. 80 and Cd Ⅱ226. 50 characteristic line was selected for calibration between the content of Cd and LIBS intensity.And PLS（ Partial Least Squares） and LSSVM（ Least Square Support Vector Machines） of themultivariate methods were applied for predicting the real content of Cd. The results demonstrated that the correlation coefficients in calibration set were 0. 9936 and 0. 9992 by PLS and LSSVM,respectively. And the correlation coefficients of validation one were 0. 9866 and 0. 9946. The values of RMSECV（ root mean square error of cross validation） were 10. 53 and 11. 59,and the RMSEP（ root mean square error of prediction） were 10. 53 and 11. 59. The average relative errors in calibration set were 17. 2% and4. 2%,and the ones in validation set were 12. 1% and 8. 6%. This work that both PLS and LSSVM can predict the Cd content in rice accurately,and LSSVM has higher reliability. This research will provide theoretical reference for heavy metal Cd detection in rice by LIBS.