特征谱区筛选在多元混合物的太赫兹光谱定量分析中的应用

用特征谱区筛选法结合太赫兹时域光谱(THz-TDS)技术对多元混合物成分含量进行了定量分析研究。实验利用太赫兹时域光谱系统测量了由乳糖一水合物(LAC)、对乙酰氨基酚(APAP)、可溶性淀粉(starch)以及微晶纤维素(MCC)四种材料组成的混合物样品的太赫兹吸收光谱，并分别尝试采用常规区间偏最小二乘(iPLS)、向后区间偏最小二乘(biPLS)、联合区间偏最小二乘(siPLS)和移动窗口偏最小二乘(mwPLS)四种特征谱区筛选法对多元混合物的太赫兹吸收光谱进行特征子区间优选，建立了太赫兹吸收谱与四元混合物中乳糖一水合物含量之间的定量回归模型。通过比较四种谱区筛选算法模型及全光谱偏最小二乘(PLS)模型所得结果，表明采用移动窗口偏最小二乘法建立的谱区筛选模型得到的结果相对最优，其交互验证均方根误差(RMSECV)、预测均方根误差(RMSEP)、校正集相关系数(R_C)和预测集相关系数(R_P)分别为0.980 3，1.114 1，0.996 0和0.995 1。实验结果表明，采用特征谱区筛选方法可以有效选择多元混合物太赫兹吸收光谱的特征区间，提高模型精度和降低模型复杂性，为实现多元混合物成分含量的快速检测提供了一种有效的方法。

Application of Interval Selection Methods in Quantitative Analysis of Multicomponent Mixtures by Terahertz Time-Domain Spectroscopy

Interval selection methods combined with terahertz time-domain spectroscopy (THz-TDS) technique were used to perform quantitative analysis of component concentrations in multicomponent mixtures. The THz spectra of 100 quaternary pharmaceutical mixtures composed of lactose monohydrate, acetaminophen, microcrystalline cellulose and soluble starch were measured using THz-TDS system. Four spectral interval selection methods, including iPLS, mwPLS, siPLS and biPLS, were employed to select spectral intervals of THz absorbance spectra of multicomponent mixtures and correlate THz absorbance spectra with the concentrations of lactose monohydrate. The mwPLS method yielded the most accurate result as compared with the other three interval selection methods and full-spectrum PLS. The optimal mwPLS model was obtained with lower root mean square error of cross-validation (RMSECV) of 0.980 3, lower root mean square error of prediction (RMSEP) of 1.114 1, higher correlation coefficient for calibration (R_C) of 0.996 0, and higher correlation coefficient for prediction (R_P) of 0.995 1. Experimental results demonstrate that spectral interval selection combined with THz-TDS could be successfully applied as an accurate and rapid method to determine component concentrations in multicomponent mixtures.