Liu Bo-ping, Qin Hua-jun, Luo Xiang, Cao Shu-wen, Wang Jun-de
College of Chemical Engineering, Nanjing University of Science and Technology, Nanjing 200014, China.
Guang Pu Xue Yu Guang Pu Fen Xi. 2007 Dec;27(12):2456-9.
The present paper introduces an application of near infrared spectroscopy (NIRS) multi-component quantitative analysis by building a kind of recurrent network (Elman) model. Elman prediction model for phenylalanine (Phe), lysine (Lys), tyrosine (Tyr) and cystine (Cys) in 45 feedstuff samples was established with good veracity. Twelve peak value data from 3 principal components straight forward compressed from the original data by PLS were taken as inputs of Elman, while 4 predictive targets as outputs. Forty seven nerve cells were taken as hidden nodes with the lowest error compared with taking 43 and 45 nerve cells. Its training iteration times was supposed to be 1000. Predictive correlation coefficients by the model are 0.960, 0.981, 0.979 and 0.952. The results show that Elman using in NIRS is a rapid, effective means for measuring Phe, Lys, Tyr and Cys in feedstuff powder, and can also be used in quantitative analysis of other samples.
本文通过构建一种递归网络（Elman）模型，介绍了近红外光谱（NIRS）多组分定量分析的一种应用。建立了45个饲料样品中苯丙氨酸（Phe）、赖氨酸（Lys）、酪氨酸（Tyr）和胱氨酸（Cys）的Elman预测模型，准确性良好。将通过偏最小二乘法（PLS）从原始数据直接压缩得到的3个主成分中的12个峰值数据作为Elman的输入，而4个预测目标作为输出。与采用43个和45个神经细胞相比，采用47个神经细胞作为隐藏节点时误差最小。其训练迭代次数设为1000次。该模型的预测相关系数分别为0.960、0.981、0.979和0.952。结果表明，Elman模型应用于近红外光谱技术是快速、有效地测定饲料粉中Phe、Lys、Tyr和Cys的手段，也可用于其他样品的定量分析。
