【摘要】 现代近红外光谱技术以其分析速度快、重现性好、成本低 ,且不消耗样品和易于实现在线分析等鲜明的特点正得到越来越多的应用。但近红外光谱严重重迭 ,需要利用化学计量学的方法建立多元回归模型对样品的物化性质进行预测。通过对建模过程和模型性能的分析 ,可实现对预测模型的优化 ,从而更好地指导未知样品的预测。文章主要介绍了在多元回归模型中具有基础作用的净信号的定义 ,并以葡萄糖的水溶液为例 ,详细说明了葡萄糖净信号的计算 ,及其与葡萄糖和水的吸光系数、样品复杂性之间的定性关系 ,最后还讨论了净信号在评价测量精度方面所起的重要作用更多还原

【Abstract】 Near-infrared spectroscopy is a rapid, high-efficiency, non-destructive and low-cost analytical technique, which has found widespread use for quantitative and qualitative analysis. Because of the complex nature of NIR spectra, multivariate calibration model plays a key role in the spectroscopic analysis. Further understandings of the model can be helpful for optimizing the model’s performance. This paper introduces a convenient definition of net analyte signal, which is one of the figures of merit charactering multivariate calibration model. The absorbance data of the glucose aqueous solution are used for the calculation of net signal and explanation of the qualitative relationships between net analyte signal of glucose and absorptivity coefficients of other components. Qualitative changes of net signal with the complexity of samples are also given. Finally a simple expression is proposed for estimating the prediction error of concentration, which is instructive for further study of the robustness and accuracy of the multivariate calibration model.更多还原