基于BP神经网络建立初烤烟叶主流烟气中氢氰酸释放量的预测模型

Establishment of Predicting Model of Hydrogen Cyanide Contents in Flue-cured Tobacco Mainstream Smoke Based on BP Neural Network

【目的】建立准确、环境友好的氢氰酸释放量预测方法,以减少检测过程中氢氰酸对人体的危害。【方法】测定了182份初烤烟叶样品的主流烟气氢氰酸释放量和25种烟叶化学成分,采用BP神经网络,以卷烟常规化学成分烟丝水分、氯、丙二酸、挥发酸、钾、总氮作为神经网络的输入,主流烟气中氢氰酸作为输出,建立初烤烟叶主流烟气中氢氰酸释放量的预测模型。【结果】利用所建模型对28个样品进行外部验证,模型平均预测相对偏差为7.88%,大部分样品的预测相对偏差在10%以内。【结论】该预测模型预测精度良好,对于初烤烟叶具有广泛的适用性。

[ Purpose] This study aimed to grasp the release of hydrogen cyanide in flue gas of mainstream tobacco and provide guidance for the adjustment of cigarette formula, finally reduce the harmfulness of cigarettes.[Method]The hydrogen cyanide（HCN） in mainstream smoke and the contents of 25 chemical components in 182 samples of cured tobacco leaves were determined. The BP neural network was applied to build the model of forecasting the HCN in mainstream smoke. The general chemical components-cut tobacco moisture, chlorine, malonic acid, volatile acid, potassium and total nitrogen were as the inputs of the network and the HCN wais as the outputs.[Results]The model was external validated by 28 samples, the average relative prediction error of HCN was 7.88% ,and the relative prediction error of most samples within 10% .[Conclusions]This model had good prediction accuracy; it could be used to flue-cured tobacco widely.