近红外光谱结合遗传算法优化的RBF神经网络对中药萃取物含量的预测

Rapid concentration determination of extractant from cnidium monnieri via near-infrared spectroscopy combined with GA-RBF neural network

建立了基于近红外漫反射光谱(NIR)定量预测中药蛇床子CO_2超临界萃取(SFE)物中2种主要成分蛇床子素和欧前胡素含量的新方法.将高效液相色谱(HPLC)分析结果作为参考值,通过遗传算法(GA)和径向基函数(RBF)神经网络相结合,建立中药蛇床子萃取物的光谱数据和萃取物中蛇床子素和欧前胡素含量之间的定量模型.NIR光谱数据经标准归一化(SNV)预处理后所建GA优化的RBF网络模型(GA-RBF)为最优,其蛇床子素和欧前胡素测试集的均方根误差(RMSE)分别为0.476 4%和0.305 6%,相关系数(R)分别为0.990 8和0.987 0,均优于偏最小二乘(PLS)模型的处理结果.该方法具有快速、无损、精确的优点,为中药材复杂体系中化学组分定量测定提供了一条新途径.

A new method for a quantitative prediction of osthol and imperatorin as active ingredients of cnidiurn monnieri obtained by means of extraction with supercritical carbon dioxide was proposed based on near-infrared （NIR） diffuse reflectance spectra. High performance liquid chromatography （HPLC） was used to determine the concentration osthol and imperatorin of cnidium monnieri for reference. A quantitative analysis model about the spectral characteristics of the cnidium monnieri extract and the content of osthol and imperatorin in the extraction was established by combining genetic algorithm （GA） with radial basis function （RBF） neural networks. The optimal network parameters and near infrared spectral region were constructed automatically by using GA. The results of the experiment were turned out that the standard normalization （SNV） was the optimal pretreatment method of modeling and showed that the root-mean-square-errors（RMSE） of osthol and imperatorin for test set were 0. 476 4% and 0. 305 6% ,respectively. The correlation coefficients（R） for test set were 0. 990 8 and 0. 987 0,respectively. That was more superior to the results of partial least squares（PLS） model. The method was fast, nondestructive and accurate. It provided a new way for quantitative determination of chemical composition in complex systems of the cnidium monnieri.