卷烟感官舒适性与主流烟气化学成分的回归建模

为更精确地拟合卷烟感官舒适性指标与烟气化学成分之间的关系，除传统的线性回归模型外，新引入三种非线性回归模型进行比较研究。同时，对检测的42种烟气化学成分采用逐步回归方法进行变量选择，以确定影响10个卷烟感官舒适性指标的关键烟气化学成分。研究结果表明:模型拟合方面在9个感官舒适性指标上非线性模型具有更好的预测性能，其中有3个指标非线性模型统计上显著优于线性模型；化学变量选择方面苯甲酸、氨和巴豆醛等3个化学成分对感官舒适性存在负面影响，甲酸等14个化学成分对感官舒适性存在正面影响，羟基乙酸、亚油酸和游离烟碱等3个化学成分对不同的感官舒适性指标有正面或负面不同的影响。研究结果可为卷烟产品研发提供科学依据。 

Regression modeling for cigarette sensory coziness and main smoke chemicals

To better fit the relationship between sensory coziness and smoke chemicals, besides classical linear regression models, three non-linear regression models were introduced to compare their performance. To identify the key chemical factors for 10 sensory coziness variables, variable selection using stepwise regression was performed on 42 measured smoke chemicals. The results show that for model fitting, non-linear models perform better than linear ones on 9 sensory variables and 3 of them, especially, are statistically significant；For variable selection, it is found that 3 chemicals, consisting of benzoic acid, ammonia and crotonic aldehyde, have negative effects to sensory coziness, 14 chemicals, consisting of methanoic acid and etc., have positive effects, and other 3 chemicals, consisting of glycolic acid, linolic acid and free nicotine, have positive or negative effects to different sensory coziness variables. The results provide a theoretical foundation for developing cigarette products.