PLS-DANN算法-NIR光谱非破坏定量分析的研究

NONDESTRUCTIVE QUANTITATIVE ANALYSIS OF COFREL IN TABLETS BY PLS-DANN ARITHMETIC COMBINED WITH NIR SPECTROSCOPY

[目的]研究偏最小二乘(PLS)结合双人工神经网络(DANN)算法-近红外(NIR)漫反射光谱非破坏定量分析方法。[方法]以Cofrel药片为研究对象,设计了最佳的PLS-DANN模型,分别讨论了最佳波长范围、导数光谱以及输入层和隐含层节点数对预测结果的影响。[结果]用HPLC法的测定结果作标准,磷酸苯丙哌林浓度预测值的相对误差RE(%)<0.14%,实现了对Cofrel药片中有效成分磷酸苯丙哌林的精确的非破坏定量测定。[结论]该方法不仅准确、可靠,而且为今后进出口商品在该领域方面的检验打下有利基础。

In this paper, a double artificial neural network （DANN） algorithm combined with partial least squares （PLS） was used to parse near infrared （NIR） reflectance spectra of Cofrel medicine in tablets. The contents of benproperine phosphate, which is effective component in the Cofrel tablets, have been accurate nondestructive quantitative predicted. The best model of PLS-DANN was designed. The effect of the best wavelength range, derivative NIR spectrum, input nodes and hidden nodes on the predicted results was discussed respectively. Compared the results with those of HPLC, the relative errors （RE%） of benproperine phosphate were less than 0.14%. The analytical results were accurate and credible, and this method could establish the foundation for the determination of similar entry-exit commodities in this field.