激光诱导击穿光谱结合CARS变量选择方法定量检测倍硫磷含量

Quantitative Determination of Fenthion Using Laser Induced Breakdown Spectroscopy with CARS Variable Selection Method

利用双脉冲激光诱导击穿光谱(LIBS)技术对溶液中的倍硫磷含量进行定量检测。采用二通道高精度光谱仪采集不同浓度倍硫磷样品在206.28~481.77 nm波段的LIBS光谱,并对光谱进行多元散射校正(MSC)、标准正态变量变换(SNV)及3点平滑预处理,根据偏最小二乘(PLS)建模确定最优的预处理方法。在此基础上,利用竞争性自适应重加权算法(CARS)筛选与倍硫磷相关的重要变量,然后应用PLS回归建立溶液中倍硫磷含量的定量分析模型,并与单变量定量分析模型及未变量选择的PLS定量分析模型进行比较。结果表明,相比单变量定量分析模型及原始光谱PLS定量分析模型,CARS-PLS定量分析模型的性能更优,其模型的校正集和预测集的决定系数及平均相对误差分别为0.969 4、15.537%和0.995 9、5.016%。此外,与原始光谱PLS模型相比,CARS-PLS模型仅使用其中1.9%的波长变量,但预测集平均误差却由9.829%下降为5.016%。由此可见,LIBS技术检测溶液中的倍硫磷含量具有一定的可行性,且CARS方法能简化定量分析模型,提高模型的预测精度。

In this study,double pulse laser induced break down spectroscopy（LIBS） technique was used to detect the content of fenthion in solution.The spectra of different concentrations’ samples in range of 206.28-481.77 nm band were collected using a two channel spectrometer with high precision.And then several preprocessing methods such as multiplicative scatter correction（MSC）,standardized normal variate（SNV） and 3 point smoothing were conducted on the spectra.The optimal pretreatment method was confirmed according to partial least square（PLS） modeling.On this basis,competitive adaptive reweighted sampling（CARS） was used to screen the important variables related to fenthion.Then,the quantitative analysis model for fenthion in solution was established by PLS regression.Finally,the CARS-PLS quantitative analysis model was compared with the single variable quantitative analysis model and the PLS quantitative analysis model without variable selection.The results indicated that the CARS-PLS quantitative analysis model has a better performance compared with single variable quantitative analysis model and PLS model,and its determination coefficient and average relative error of calibration set and prediction set are 0.969 4,15.537% and 0.995 9,5.016%,respectively.Furthermore,the CARS-PLS model adopted only 1.9% of the wavelength variables,but the average error of the prediction set was decreased from 9.829% to 5.016%.Thus it is found that LIBS technology has a certain feasibility to detect the content of fenthion in solution.And CARS could simplify the quantitative analysis model and improve the prediction accuracy of the model.