烤烟颜色特征值对卷烟主流烟气有害成分影响的BP神经网络构建

Construction of BP neural network for the influence of flue-cured tobacco color characteristic value on harmful components of cigarette mainstream smoke

为研究初烤烟叶颜色特征值对烟气有害成分的影响,对南阳市属县(市)的174个中部叶C3F等级的烟叶样品进行颜色特征值、卷烟主流烟气有害成分测定,建立BP神经网络模型,对卷烟主流烟气中的有害成分进行分析。结果表明,因分析的6个主因子的累计贡献率达到92.2%,所构建的BP神经网络模型训练样本的真实值与预测值的决定系数均达到了0.97以上,验证样本的主流烟气中CO真实值与预测值的决定系数达到0.9342外,其余6种有害成分真实值与预测值的决定系数均达到了0.96以上。由此可见,所建立的网络模型的鲁棒性较好,能够在一定程度上用烤烟的颜色特征值来表征卷烟主流烟气中的有害成分含量。

In order to study the influence of the color characteristic value of flue-cured tobacco leaves on the harmful components of flue gas,the color characteristic values and the harmful components of cigarette mainstream smoke of the 174 middle leaf C3F tobacco samples in Nanyang subordinate counties and cities were determined,and BP neural network model was established to analyze the harmful components in the mainstream smoke of cigarettes.The results show that the cumulative contribution rate of the six analyzed principal factors was 92.2%,so this is used as a representative factor.The determined coefficient of the real and predicted values of the BP neural network model training samples was above 0.97.The coefficient of determination of the true and predicted values of CO in the mainstream smoke of the sample was 0.9342.The coefficient of determination of the real and predicted values both above 0.96.It can be seen that the established network model is robust,and the color characteristic value of flue-cured tobacco can be used to some extent to characterize the harmful components in the mainstream extension of cigarettes.