人工神经网络紫外光度法同时测定甲基苯甲醛的三种同分异构体及其与偏最小二乘法的比较

同分异构体的同时测定一直是分析化学领域的热点和难点问题，本文将化学计量学中的多元校正方法，如偏最小二乘法和人工神经网络法与紫外分光光度法相结合，同时测定了紫外吸收光谱严重重叠的甲基苯甲醛的三种同分异构体混合体系中各组分的含量。确定了测定的最佳波长范围为230～304 nm；测得48个混合标样的吸光度值用于建立模型，其中，邻、间、对甲基苯甲醛的浓度范围分别为6.0～15.0、7.0～16.0和8.0～19.0 μg·mL-1。7个模拟样品作为监测集用于检验所建立模型的预测性能。本文还讨论了三种组分浓度比例对所建立模型预测性能的影响并确定了可以准确测定的浓度比例范围。所建立的方法用于模拟样品的测定，其回收率在84.00%与109.60%之间。与偏最小二乘法的测定结果比较，经成对t检验表明，两种方法对邻、间甲基苯甲醛测定结果无显著性差异；而对甲基苯甲醛的测定，人工神经网络法的测定结果优于偏最小二乘法。

Simultaneous Spectrophotometric Determination of Three Tolualdehyde Isomers by Artificial Neural Networks and Its Comparison with Partial Least Squares

The simultaneous determination of tolualdehyde isomer mixtures by using a spectrophotometric method is a difficult problem in analytical chemistry, due to spectral interference. By multivariate calibration methods, such as partial least squares (PLS) and artificial neural network (ANN), it is possible to obtain a model adjusted to the concentration values of the mixtures used in the calibration range. The ANN calibration model of simultaneous determination of tolualdehyde isomers was based on absorption spectra in the 230–304 nm range for 48 different mixtures of tolualdehyde isomers, and the concentration ranges for o‐tolualdehyde, m‐tolualdehyde and p‐tolualdehyde were 6.0–15.0, 7.0–16.0 and 8.0–19.0 µg·mL^?1, respectively. Other 7 samples were applied to testing the predictive ability of the calibration model. The effect of the three isomers with different concentration ratios in their mixtures on the accuracy of determination results was discussed and the appropriate range of concentration ratio was found. The proposed method was applied to simultaneous determination of the three isomers in synthetic samples and the recoveries were between 84.00% and 109.60%. The results of determination were compared with the PLS model without significant difference at 0.05 level for o‐tolualdehyde and m‐tolualdehyde by paired t‐test. But a better performance for p‐tolualdehyde in their mixtures could be obtained by the ANN calibration model.