主成分回归法建立卷烟风格与组分的定量关系

为研究卷烟风格与卷烟化学成分之间的关系,应用主成分回归建立卷烟GC-MS数据与卷烟风格评吸值之间的定量模型.对于每种卷烟风格,均建立了4种候选模型,然后选取最佳者.通过最佳模型对测试集样品的30种风格进行预测,有23种风格的预测值与评吸值之间的平均绝对误差小于专家评吸时的最小计分单位,所以定量模型的预测结果可靠.此研究表明化学计量学方法在处理卷烟这种复杂体系时的可用性和有效性.这些定量模型可以作为专家评吸卷烟风格的辅助工具.在定量关系不明确的情况下,应当建立多个候选模型,然后从中选择最佳者.

Building Models Using Principal Component Regression to Analyze the Relationship Between Tobacco Flavors and the Constituents

The relationship between tobacco flavors and their chemical constituents was investigated. GC-MS data of tobacco samples were measured,and quantitative models between tobacco flavors and GC-MS data were built with principal component regression method. For each flavor,four candidate models were built and the optimal one with minimum mean absolute value was chosen. Mean absolute error of 23 flavors obtained from the optimal models are smaller than the precision of flavor evaluations by professionals, which means the predictions are reliable. The models can be used as auxiliary tools for tobacco rating professionals. The availability and effectiveness of chemometric methods are demonstrated in this study. It is also demonstrated that in processing complex systems such as tobacco,it is desirable to build several models and choose the optimal one.