多元线性回归提高激光诱导荧光辅助激光诱导击穿光谱技术的准确度

冶金、核工业、污染检测和环境监测等领域对元素分析的需求是必不可少。激光诱导击穿光谱技术作为一种新型的原子光谱分析技术，具有实时快速、对样品几乎无损、可多元素同时分析等特点，因此一直受到广泛的关注。但其分析灵敏度较差的缺点一直限制着该技术的发展。激光诱导荧光辅助激光诱导激光光谱技术能够通过激光共振激发提高分析灵敏度并高效检测样本元素种类，通过光谱仪收集光谱信息并建立模型可对未知样本进行浓度预测。但当基体原子与目标原子的特征谱线十分接近时，基体谱线会受到影响，此时一元定标准确度下降。通过一元线性拟合和多元线性拟合两种方式对钢铁中的Ni和Cr元素分别建立线性模型。首先，选取样品光谱中的峰值谱线，核实其是否为待测元素或基体元素所对应的特征谱线，选定合适的特征谱线后，将多个谱线的光谱强度以及对应该样品的待测元素浓度作多元线性拟合模型，将各个谱线所对应的拟合系数由高到低进行排序，并以多元线性拟合模型中各个特征谱线对应的光谱强度对浓度预测的贡献度为标准不断减少拟合维度，使Ni和Cr拟合模型的决定系数分别由0.960 1提高至0.992 9和0.992 0提高至0.998 7，Ni和Cr元素含量的回归模型平均相对误差分别由38%降低至10%左右和55%降低至25%以内，Ni和Cr元素的线性回归模型的交叉验证均方根误差随着维度的增加分别由3.4%降低至2%左右和2.5%降低至1.5%左右。选取多个谱线建立多元线性回归模型的方法较为有效的降低了激发干扰的影响，以较小的工作量提高了对待测样品的待测元素浓度预测的准确度，为推进激光诱导荧光辅助激光诱导激光光谱技术在元素分析的实际应用提出了一种可行的方案。

Multiple Liner Regression for Improving the Accuracy of Laser-Induced Breakdown Spectroscopy Assisted With Laser-Induced Fluorescence (LIBS-LIF )

Elemental analysis is an essential requirement in the metallurgical industry, nuclear industry, pollution detection and environmental monitoring. As a new type of atomic spectrum analysis technology, LIBS has been widely concerned because of its real-time, fast, almost non-destructive and multi-element simultaneous analysis. However, its poor analytical sensitivity has restricted the development of this technology. LIBS-LIF can improve the sensitivity of analysis and efficiently detect the element types of samples through laser resonance excitation. The spectrometer can collect spectral information and a model can be established to predict the concentration of unknown samples. However, when the characteristic spectral lines of the matrix atom and the target atom are very close, the matrix spectral lines will be affected, and the unary calibration accuracy will decrease. In this paper, linear models of Ni and Cr elements in steel were established using linear fitting with one variable and linear fitting with multiple variables. Firstly, the peak spectral line in the sample spectral map is selected to find whether it is the characteristic spectral line corresponding to the element to be measured or the collective element. After selecting suitable characteristic spectral lines, the spectral intensities of multiple spectral lines and the concentrations of the elements to be measured in the sample were used as a multivariate linear fitting model, and the fitting coefficients corresponding to each spectral line were ranked from highest to lowest, and the contribution of the spectral intensities corresponding to each characteristic spectral line in the multivariate linear fitting model to the concentration prediction was taken as the criterion from highest to lowest, and the fitting dimension was increased continuously. The mean relative errors of the regression models for Ni and Cr elemental content were reduced from 38% to about 10% and 55% to within 25%, respectively, and the root mean square error values of the cross-validation of the linear regression models for Ni and Cr elemental content were reduced from 3.4% to 2% and 2.5%, respectively, with the increase of dimensionality. and 2.5% to 1.5% for Ni and Cr, respectively. In this paper, the method of selecting multiple spectral lines to establish a multiple linear regression model is relatively effective in reducing the influence of excitation interference, and it puts forward a feasible scheme for promoting the practical application of laser-induced fluorescence assisted laser-induced laser spectroscopy technology in element analysis.