基于PLS的水体重金属LIBS特征变量筛选方法研究

发布时间：2018-06-20 18:24

  本文选题：光谱学 + 激光诱导击穿光谱　； 参考：《光谱学与光谱分析》2017年08期

【摘要】：在水体重金属激光诱导等离子体光谱定量分析中,一般提取光谱的多个特征变量进行浓度反演,但变量之间所包含的光谱信息可能存在重叠,回归模型的复杂程度也随之增大。为提取有效特征变量,研究了基于偏最小二乘法(PLS)的变量筛选方法。该方法以待测元素浓度为因变量,多个与待测元素浓度相关的LIBS光谱特征值为自变量,进行PLS建模;依据各原始变量的投影重要性指标值进行变量筛选,提取最优变量子集。结果表明湖库水体中Pb元素的最优变量子集为PbⅠ405.78nm峰值及峰值前相邻点光谱值、内标校正值和信背比值,训练集的复相关系数R2m=0.912。以优化变量组合进行PLS回归分析,测试集预测结果的RSD和RE分别为10.2%和7.9%,显著优于内标法的预测结果。结果还表明,变量筛选结果对于不同元素和不同水样具有一定适用性。研究结果为水体重金属LIBS定量分析提供了优质特征数据,研究方法为其他涉及变量筛选的定量分析提供了参考。
[Abstract]:In the quantitative analysis of laser induced plasma spectra of heavy metals in water, the concentration inversion is carried out by extracting several characteristic variables of the spectra, but the spectral information contained among the variables may overlap, and the complexity of the regression model also increases. In order to extract effective feature variables, the method of variable selection based on partial least square method (PLS) was studied. This method takes the concentration of tested elements as dependent variables and several LIBS spectral eigenvalues related to the concentration of tested elements as independent variables to model PLS model, and selects variables according to the projection importance index values of each original variable to extract the optimal subset of variables. The results show that the optimal subset of Pb elements in lake and reservoir water is the peak value of Pb 鈪，

本文编号：2045265