摘要
为快速检测贝类重金属的污染,采用激光诱导击穿光谱(LIBS)结合多模型共识策略来预测泥蚶重金属铜的含量.分析铜元素的LIBS特征谱线,提取铜元素的特征峰强度、面积、峰强比等变量,分别构建多元线性回归模型,以及基于铜特征光谱区域信息构建偏最小二乘模型.在这些校正的成员模型预测残差向量间,通过拉格朗日乘数法优化各成员模型的线性组合,使共识模型的均方误差和最小,从而获得各成员模型的最优权系数.经外部预测集的验证,共识模型的预测结果优于任一成员模型,预测均方根误差为20.641mg/kg,相关系数为0.835,且预测偏差仅为-0.473,表明LIBS技术结合共识模型能用于重金属的定量检测.
In order to rapidly detect the heavy metal pollution in shellfish, the contents of heavy metal copper in tegillarca granosa were predicted by Laser-induced Breakdown Spectroscopy （LIBS） combined with the consensus strategy of mulit models. By analysis of LIBS copper characteristic spectral lines, the characteristics of copper peak intensity, area, peak ratio of copper＇s peaks, were extracted, and multivariate linear regression models were developed respectively. Besides, partial least regression model was calibrated based on copper characteristic spectra region. Among the residual vectors of these calibration member models, the Lagrange multiplier method was used to optimize the linear combination between these member models, aiming at reducing the correlation between member models and minimizing the mean squared error of the consensus model. Through external validation from the prediction set, the consensus model was performed better than that of any member models, with the root mean square prediction error of 20.641 mg/kg, as well as the correlation coefficient of 0.835, and the prediction deviation was only -0. 473. Results show that the LIBS technology combined with the consensus model can be used for the quantitative detection of heavy metals of aquatic products.
作者
郭珍珠
陈孝敬
袁雷明
陈熙
朱德华
杨硕
GUO Zhen-zhu;CHEN Xiao-jing;YUAN Lei-ming;CHEN Xi;ZHU De-hua;YANG Shuo(College of Mathematics,Physics and Electronic Information Engineering,Wenzhou University,Wenzhou,Zhejiang 325035,China)
出处
《光子学报》
EI
CAS
CSCD
北大核心
2018年第8期106-111,共6页
Acta Photonica Sinica
基金
国家自然科学基金(Nos.61705168
31571920)
温州市公益计划项目(No.S20170003)资助~~
关键词
激光诱导击穿光谱
共识模型
水产品
重金属
定量检测
Laser induced breakdown spectroscopy
Consensus model
Aquatic product
Heavy metals
Quantitative detection
