一种基于局部回归算法的多组分定量分析方法

文章针对复杂样本吸收光谱重叠或组分之间相互作用导致偏离朗伯一比尔定律的问题提出了一种以灰色综合关联度作为样本相似性判据的多组分定量分析局部回归建模方法，主要内容是对校正集样本的光谱曲线与待测样本曲线进行灰色综合关联度分析，然后以最小预测均方根误差原则选择与待测样本属性相近的样本组成校正子集，最后建立基于校正子集的偏最小二乘回归模型。相比马氏距离方法，灰色综合关联度结合了绝对位置差和变化率两方面因素，能够更为全面的反映样本之间的相似程度。建立实验系统将本方法应用于食用色素苋菜红、胭脂红、柠檬黄和日落黄混合溶液的定量分析中，实验结果表明，该方法优于全局建模方法，尤其在光谱响应与浓度之间的非线性响应段预测精度得到了明显的提升。

A Local Regression Method Based on Multicomponent Quantitative Analysis

resolve the problem of deviations from the Lambert-Beer law caused by overlapping spectral bands or interactions between components, a new local regression method of muhicomponent quantitative analysis based on the synthesized grey correlation is proposed. Firstly, the synthetic degree of grey relation coefficient between the spectrum of the sample to be predicted and each of the sample spectra in the calibration set was calculated. Secondly, the calibration subset was built for each prediction sample according to principle of least root mean square error of prediction. Samplesdisplaying the higher correlation are selected to constitute calibration subset of homogenous physical and chemical attributes. Finally, the multivariate calibration model was constructed based on partial least squares regression with calibration subset for each prediction sample. Compared with the existed Mahalanobis distance method, the proposed method can completely describe the similarity of the samples both in absolute deviation and rates of change. The experiment measurements are demonstrated by applying the method to the simultaneous determination yellow FCF samples and experimental results show that of amaranth, carmine, lemon yellow and sunset the method has a significant improvement in the precision compared with Global regression method, especially in the situation of non-linear relationship between concentration and spectroscopic data of the samples with a great concentration difference.