为通过烟草香味成分评价其内在品质,采用水蒸气蒸馏法-气质联用法分析了38种烟叶样品的香味成分,评吸了其单料卷烟.并以30个烟叶样品作训练样本,8个样品作预测样本,采用遗传算法GA-BP神经网络法建立了烟草香味成分分析数据与其评吸得分的预测模型.结果表明:烟草样品中共鉴定出76种香味成分;通过GA法选择出28种与烟叶评吸总分显著相关的成分,由这些成分建立的GA-BP神经网络模型,其训练样本拟合误差＜2%,预测误差＜5%.%In order to evaluate the smoking quality of tobacco leaves by their aroma components, the aroma components in 38 tobacco leaf samples were determined by steam distillation and GC/MS, and the sensory quality of cigarettes made of each tested tobacco was evaluated by panel test.A predicting model was developed by the data of aroma components and sensory scores of tobacco samples with genetic algorithm (GA)-BP neural networks using 30 samples as a training set, and the other 8 samples served as a predicting set.The results showed that 76 aroma components were identified in the samples, of which 28 components significantly correlated with sensory score were selected by GA, the fitting errors of the established model were less than 2％ and its prediction errors were within 5％.
展开▼
