卷烟主流烟气成分释放量的人工神经网络预测研究

Predication studies of mainstream cigarette smoke constituents by BP-ANN

应用标准方法测试了55种烤烟样品(中国43种、巴西6种和津巴布韦6种)常规化学成分：总糖、还原糖、总植物碱、总挥发碱、总氮、钾、氯和卷烟主流烟气化学成分：烟碱、焦油、一氧化碳。完成了烟草淀粉、石油醚提取物以及卷烟主流烟气苯并[a]芘的分析。通过误差反向传播网络(BP-ANN)对卷烟主流烟气成分释放量进行了预测研究。结果显示,除烟气烟碱、苯并[a]芘的相对偏差绝对值的平均值略超过10%外,焦油、一氧化碳和抽吸口数的相对偏差平均值均小于10%,说明应用BP网络可通过烟草化学成分预测卷烟主流烟气成分烟碱、焦油和一氧化碳的释放量。

The routine chemical components of fifty-five kinds of flue-cured tobacco samples （43 tobacco samples from,China, 6 from Brazil and 6 from Zimbabwe） , including total sugars, reducing sugars, total alkaloids, total volatile bases, total nitrogen, potassium and chlorine, were determined with standard methods. Nicotine, tar and carbon monoxide of mainstream cigarette smoke were also analyzed with standard methods. Starch, petroleum ether extract of tobacco and benzol a] pyrene of mainstream cigarette smoke were also analyzed. Cigarette smoke constituents produced from tobacco combustion were forecasted by BP-ANN. The preeisions of BP-ANN were discussed to predicate mainstream cigarette smoke constituents from tobacco variables. The averages of relative deviation of tar, carbon monoxide and puff number are less than 10 percent except nicotine and B[a]P. The BP-ANN may be applied to forecast nicotine, tar and carbon monoxide of mainstream cigarette smoke from tobacco components.