基于内标法和CARS变量优选的倍硫磷含量LIBS检测

利用共线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研究。采用石墨对倍硫磷溶液进行富集，利用双通道高精度光谱仪获取样品的LIBS光谱。以碳元素谱线（CⅠ247.856 nm）为内标对210~260 nm波段谱线进行校正，然后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波长变量，最后应用最小二乘支持向量机（LSSVM）建立倍硫磷含量的定标模型，并与基本定标法及内标法建立的单变量定标模型进行比较。研究结果表明，共线双脉冲LIBS技术可以用于溶液中的倍硫磷含量检测。基本定标法建立的最优定标模型的拟合度R²为0.935 04，预测集样品的平均预测相对误差（PRE）为41.50%；内标法建立的最优单变量定标模型的拟合度R²为0.993 61，预测集样品的平均PRE为14.91%；内标-CARS-LSSVM定标模型的拟合度R²为0.998 6，预测集样品的平均PRE为8.06%。对比上述3类定标模型，内标-CARS-LSSVM定标模型性能最优，内标法建立的定标模型次之，而基本定标法建立的定标模型最差。由此可知，CARS方法可以有效筛选倍硫磷相关的重要变量，内标法结合CARS及LSSVM方法可以改善定标模型性能，提高预测精度。

Detection of Fenthion Content by LIBS Combined with Internal Standard and CARS Variable Selection Method

The col-linear double pulse laser induced breakdown spectroscopy(Laser induced breakdown spectroscopy, LIBS) technology was used to detect fenthion in solution. The fenthion solution was enriched with graphite. The LIBS spectra of samples were obtained by double channel high-precision spectrometer. The carbon spectral line(C Ⅰ 247.856 nm) was used as internal standard to adjust spectral line of 210-260 nm. Then, the competitive adaptive reweighted sampling(Competitive adaptive reweighted sampling, CARS) was used to screen important variable wavelengths which related to fenthion. Finally, the least squares support vector machine(Least squares support vector machine, LSSVM) was used to establish the calibration model of fenthion, and the model was compared with the standard calibration model which established by the basic calibration method and the internal standard method. The results show that the col-linear double pulse LIBS technique can be used to detect the content of fenthion in solution. Optimum calibration model's fitting degree R² of basic calibration method is 0.935 04. The prediction relative error(Prediction relative error, PRE) of prediction samples is 41.50%. While fitting degree R² of internal standarad method is 0.993 61, and PRE of prediction samples is 14.91%. What's more, fitting degree R² of internal standard and CARS-LSSVM method is 0.998 6, and PRE of prediction samples is 8.06%. Comparing the three types of calibration models, it can be found that the internal standard and CARS-LSSVM calibration model has the best performance. The calibration model which established by internal standard method is second. And the calibration model which established by the basic calibration method is the worst. It can be seen that the CARS method can effectively screen the important variables which related to fenthion. Internal standard method combined with CARS and LSSVM methods can improve calibration model's performance and improve the accuracy of predicting.