便携式X射线荧光光谱法结合支持向量回归算法定量分析土壤中的砷含量

研究了基于统计学习理论的支持向量机（SVM）回归法在X射线荧光光谱定量分析中的应用。以39个农田土壤样品作为实验材料，以其中32个土壤样品作为校正集，选用SVM模型中Linear、Poly和RBF 3种核函数对As元素含量与荧光光谱数据进行回归建模。用3种不同模型对预测集中7个土壤样品的As元素含量进行预测分析，结果显示模型预测As元素含量与电感耦合等离子体发射光谱法测定的As元素含量之间的相关系数R2均大于0.99，相对分析误差RPD均大于3，表明所建立的SVM模型具有较好的使用价值。为了进一步考察SVM回归模型的预测效果，同应用较成熟的PLS回归模型的预测结果进行对比，结果显示SVM法的预测结果更好，表明SVM回归模型亦可用于便携式X射线荧光光谱法的定量预测分析。

Quantitative analysis of arsenic in soil by portable fluorescence spectrometer combined with support vector machine regression

Application of Support Vector Machine (SVM) Regression Method Based on Statistical Learning Theory in Quantitative Analysis of X-ray Fluorescence Spectrum was studied. 39 farmland soil samples were used as experimental materials, of which 32 soil samples were used as correction sets. Three kernel functions linear, poly and RBF in SVM model were used to model the regression between As content and fluorescence spectrum data. Three different models were used to predict and analyze the As content of seven soil samples in the prediction set. The results show that the correlation coefficient R2 between the As content predicted by the model and the As content determined by inductively coupled plasma emission spectrometry is greater than 0.99, and the relative analysis error RPD is greater than 3, indicating that the SVM model has good application value. In order to further investigate the prediction effect of SVM regression model, compared with the prediction results of more mature PLS regression model, the results show that the prediction results of SVM method are better, indicating that SVM regression model can also be used for quantitative prediction and analysis of portable X-ray fluorescence spectrometry.