近红外光谱法快速检测卷烟纸中的多种离子
Rapid detection of various ions in cigarette paper by near-infrared spectroscopy
-
摘要:
为快速分析卷烟纸中多种离子含量，应用傅里叶变换近红外光谱（FT-NIR）分析技术和常规化学分析方法分别测定了202个卷烟纸样品的光谱数据和钾、钠、钙、镁、氯、硝酸根、硫酸根、磷酸根、柠檬酸根及酒石酸根等多种离子含量。利用测得的近红外光谱数据和化学值，结合偏最小二乘回归算法（Partial least squares regression， PLSR）和最小二乘支持向量机回归算法（Least squares support vector regression， LS-SVR），建立了卷烟纸中多种离子的数学预测模型。结果表明：采用PLSR算法建立的卷烟纸中的化学成分模型中，仅有镁离子和氯离子能满足快速检测的需求（模型拟合系数均在0.90以上，预测误差较小）。采用LS-SVR算法建立的化学成分模型的效果均优于PLSR模型。对于钾、钠、镁、氯离子和柠檬酸根等离子，LS-SVR模型中测试集的决定系数均为0.90以上；对于钙离子，LS-SVR模型中测试集的决定系数为0.85，且预测误差较小。说明上述6种离子的模型可用于卷烟纸的批量快速检测；对于硫酸根，尽管模型的拟合能力可以接受（R2=0.72），但是由于模型预测值和实际值之间的误差达到25%以上，不能应用于快速检测；而对于硝酸根、酒石酸根和磷酸根离子，模型拟合能力和预测误差均不能接受，不能用于卷烟纸中对应化学成分的检测。 Abstract:In order to rapidly analyze contents of various ions in cigarette paper, Fourier transform near-infrared spectroscopy (FT-NIR) and traditional chemical analytical method were employed to collect spectral data and the contents of chemical components (K+, Na+, Ca2+, Mg2+, Cl－, NO3－, SO42－, PO43－, citrate ion, and tartrate ion) of 202 cigarette paper samples. Combined with partial least squares regression (PLSR) and least squares support vector regression (LS-SVR) algorithms, mathematical prediction models for various ions in cigarette paper were established based on the spectral data and chemical values. The results showed that, among the models based on the classical PLSR algorithm for chemical components in cigarette papers, only the models of Mg2+ and Cl－ could meet the needs of rapid detection (their model fitting coefficients were all larger than 0.90, their prediction errors were smaller). Models based on LS-SVR algorithm for chemical components performed better than PLSR models. Among them, the R2 of the test set in the LS-SVR models for K+, Na+, Mg2+, Cl－ and citrate ion were all more than 0.90, while for Ca2+, the R2 of the test set in the LS-SVR model was 0.85, and the prediction error was smaller, which demonstrated that the models for the above-mentioned six ions could be used in the rapid detection for cigarette paper. While for SO42－, although the R2 (R2=0.72) could be accepted, the model could not be used for rapid detection because the error between predicted value and real value was over 25%. For the other three anions, both fitting abilities and prediction errors were unacceptable, therefore the models could not be used for the predictions of contents of NO3－, PO43－ and tartrate ion.
下载: