论文部分内容阅读
为研究烤烟香型风格的定量评价方法,利用主成分分析和Back Propagation(BP)神经网络法构建了烤烟香型与烟叶化学成分指标之间的预测模型。结果表明:糖碱比、碱氮比、总烟碱、总糖、氯含量5种化学成分指标对烤烟香型的累计贡献率达97.87%,以这5种化学成分构建的BP神经网络模型对烤烟香型的预测结果与人工评吸结果的总吻合率达到89.36%,其中清香、浓透清、浓偏中等香型预测吻合率达到100%。这说明利用非线性模型来评定或预测烤烟香型风格特征是可行的。
In order to study the quantitative evaluation method of the flue-cured tobacco flavor, a prediction model between the flue-cured tobacco flavor and the tobacco chemical composition index was constructed by using principal component analysis and Back Propagation (BP) neural network. The results showed that the cumulative contribution rate of five chemical components, including ratio of nicotine to base, ratio of total to total nicotine, total nicotine, total sugar, and chlorine, to flue-cured tobacco was 97.87%. BP neural network model constructed from these five chemical components The total coincidence rate of flue-cured tobacco flavor prediction results with artificial smoking results reached 89.36%, of which the coincidence rate of the fragrances, the thicker clear and the thicker ones was 100%. This shows that it is feasible to use non-linear models to evaluate or predict the flavor characteristics of flue-cured tobacco.