为解决多元分析方法建模成本高及适用性差, 严重制约其在光谱分析领域大量应用的难题, 提出了一种新的多元定量分析方法。 以汽油及其甲基叔丁基醚(MTBE)溶液的红外光谱为研究对象, 使用五种汽油及其窄馏分建立背景光谱空间, 用斜投影算法分离出混合光谱中纯MTBE光谱分量, 建立纯光谱响应值与浓度之间的标准曲线, 相关系数为0.995 2, 截距仅为0.025, 实现了多组分复杂体系中待测物的定量分析。 与正交投影方法比较, 新方法对五个预测样本的预测结果明显优于正交投影法。 对17个实际油样的预测结果显示, 新方法比PLS模型具有更好的适用性。 新方法无需收集大量样本和建立复杂模型, 方法简单, 准确, 适用性好。

In the spectral analysis, a large-scale application of the traditional multivariate analysis methods has been limited by both high cost and poor applicability of the calibration models. A new multivariate analysis method was proposed for multicomponent systems in the present paper. Determining MTBE content in gasoline solution by infrared spectroscopy was studied. The spectra of five kinds of gasoline and their 50 narrow distillation fractions were used to build the background library. The oblique projection algorithm was applied to the spectra of MTBE gasoline solution samples to extract the pure spectral signal of MTBE in the solution. A unary linear regression curve was built between the pure spectral signals of MTBE and their concentrations with a correlation coefficient of 0.995 2 and an intercept of 0.025. Compared with the orthogonal projection algorithm method and PLS model method, a large amounts of calibration samples and complex model are no longer needed by the new method which is simpler, more accurate and with better applicability.