X-ray fluorescence spectra quantitative analysis based on characteristic spectra optimization of partial least-squares method  

作　　者：章炜 段连飞 张罗政 张玉钧 凌六一 杨允军 

机构地区：[1]New Star Research Institute of Applied Technology, Hefei 230031, China [2]Key Laboratory of Environment Optics and Technology, Anhui Institute of Optics ~nd Fine Mechanics, the Chinese Academy of Sciences, Hefei 230031, Chin~

出　　处：《Chinese Optics Letters》2014年第A02期144-148,共5页中国光学快报（英文版）

摘　　要：The quantitative analysis of X-ray fluorescence （XRF） spectra is studied using the partial least-squares （PLS） method. The characteristic variables of spectra matrix of PLS are optimized by genetic algorithm. The subset of multi-component characteristic spectra matrix is established which is corresponding to their concentration. The individual fitness is calculated which combines the crossover validation parameters （prediction error square summation） and correlation coefficients （R^2）. The experimental result indicates that the predicated values improve using the PLS model of characteristic spectra optimization. Compared to the nonoptimized XRF spectra, the linear dependence of processed spectra averagely decreases by about 7%, root mean square error of calibration averagely increases by about 79.32, and root mean square error of cross-validation avera^elv increases by about 14.2.The quantitative analysis of X-ray fluorescence （XRF） spectra is studied using the partial least-squares （PLS） method. The characteristic variables of spectra matrix of PLS are optimized by genetic algorithm. The subset of multi-component characteristic spectra matrix is established which is corresponding to their concentration. The individual fitness is calculated which combines the crossover validation parameters （prediction error square summation） and correlation coefficients （R^2）. The experimental result indicates that the predicated values improve using the PLS model of characteristic spectra optimization. Compared to the nonoptimized XRF spectra, the linear dependence of processed spectra averagely decreases by about 7%, root mean square error of calibration averagely increases by about 79.32, and root mean square error of cross-validation avera^elv increases by about 14.2.

分 类 号：O657.3[理学—分析化学] O657.34[理学—化学]