LM-BP神经网络应用于光度法同时测定邻苯二酚、间苯二酚和对苯二酚三组分

应用人工神经网络原理，以Levenberg-marquardt back propagation（LM-BP）算法，对光谱严重重叠的邻苯二酚、问苯二酚和对苯二酚3组分进行同时测定。挑选283．5，279．5，276．5nm处3波长为3种苯二酚的测量波长，按正交设计表L25（5^6）配置25组标准混合溶液，对3种苯二酚进行训练，平均平方误差（MSE）达到最小值0．0831143；同时预测了模拟混合样本中的邻苯二酚、间苯二酚和对苯二酚含量，3种苯二酚异构体在低浓度时误差稍大，绝大部分分析结果相对误差小于5％，特别是对吸收光谱严重重叠的邻苯二酚、间苯二酚有令人满意的分析结果。

Spectrophotometric Simultaneous Determination of Pyrocatechol, Resorcinol and Hydroquinone by LM-BP Neural Network

By means of artificial neural network and Levenberg Marquardt Back Propagation (LM-BP) train algorithm, the three components of pyrocatechol, resorcinol and hydroquinone were determined simultaneously. The absorption spectra of these three components severely overlap in ultraviolet spectral range. Three wavelengths at 283.5, 279.5 and 276.5 run were selected for the determination. 25 mixture standard solutions were prepared according to orthogonal projection form L25 (56 ). Three kinds of components were trained. Mean Squared Error (MSE) reaching minimum value is 0.083 114 3. Meanwhile the contents of pyrocatechol, resorcinol and hydroquinone in six simulation mixture samples were predicted. The relative errors of the three kinds of components were slightly larger under the low concentration condition, and the mean relative error for most analytical results was less than 5 % , especialmy it is satisfactory for the analytical results of pyrocatechol and resorcinol with severely overlapped absorption spectra.