最小二乘支持向量机和内标法的乐果农药含量LIBS检测

利用共线双脉冲激光诱导击穿光谱 (LIBS)对溶液中的乐果含量进行定量检测。采用圆柱形桐木木片对农药乐果进行富集，然后利用双通道高精度光谱仪获取样本在206.28~481.77 nm波段范围的LIBS光谱。选用4条磷元素谱线(P Ⅰ 213.618 nm，P Ⅰ 214.91 nm，P Ⅰ 253.56 nm，P Ⅰ 255.325 nm)为分析线，碳元素谱线(C Ⅰ 247.856 nm)为内标线，应用单变量线性拟合及最小二乘支持向量机(LSSVM)方法分别建立溶液中乐果含量的单变量定标模型、LSSVM定标模型及基于内标法的LSSVM定标模型，并进行比较。三个定标模型中，基于内标法的LSSVM定标模型性能最优，LSSVM定标模型性能次之，而单变量定标模型性能最差。结果表明，共线双脉冲LIBS技术结合LSSVM及内标法可以用于溶液中的乐果含量定量检测，所建立的定标模型的决定系数为0.999 7，训练集和验证集的平均相对误差分别为11.24%及12.01%。LSSVM方法及内标法均能在一定程度上改善定标模型的性能，提高预测精度。

Detection of Dimethoate Content with Laser Induced Breakdown Spectroscopy Combined with LSSVM and Internal Standard Method

In this research, collineardouble pulselaser induced breakdown spectroscopy (LIBS) was used to detect dimethoate content in solutionquantificationally. Fortune paulownia wood chip with cylinder shape was used to enrichmentdimethoate, and the spectra of samples were acquired with a two-channel high precision spectrometer in the wavelength range of 206.28~481.77 nm. Four spectral linesof phosphorus (213.618, 214.91, 253.56, 255.325 nm) were selected as analytical lines, and carbonspectral line (247.856 nm) was used as internal standard line. Then, univariatelinear fitting and least squares support vector machine (LSSVM) were used to develop univariate calibration model, LSSVM calibration model and LSSVM calibration model based on internal standard method, and the performance of threecalibration models were compared. The results indicate that collinear double pulse LIBS combined with LSSVM and internal standard method is feasible for detecting dimethoate content in solution quantificationally. The coefficient of determination (R2) of LSSVM calibration model based on internal standard method is 0.999 7, and the average relative errors in training set and validation set are 11.24% and 12.01%, respectively. In the three calibration models, LSSVM calibration model based on internal standard method has the best performance, and the performance of LSSVM calibration model is the second, while univariatecalibration model hasthe worstperformance. So it can be concluded that LSSVM and internal standard method can improve the performance of calibration model to some extent, and improve the prediction accuracy.