In order to explore the relationship between the intrinsic chemical composition and the cigarette smoke index and the sensory quality score,the model of the neural network model for predicting cigarette smoke and sensory quality is established.Prediction of flue gas components and sensory score by routine chemical compositions based on the BP neural network of A brand.The hidden layer node is 9,the input function is Tansig,the output function is purelin.Training method for gradient descent method.22 samples are selected as training samples,19 samples as the validation sample and 3 samples as the test sample.The goal of the training is to allow the error is 10 000,the maximum number of iterations is 0.000 1 times.The predicted results are compared with the conventional chemical detection and the actual results.The relative standard deviation is less than 5％.The model has the guiding significance for predicting the release quantity and the sensory evaluation of the cigarette mainstream smoke components.%为了探索内在化学成分与卷烟烟气指标和感官品质得分之间的关系,建立相应的预测卷烟烟气指标和感官品质得分神经网络模型数学模型.测试了A牌号卷烟不同批次成品卷烟常规化学成分、主流烟气化学成分和感官得分,以常规化学成分作为网络输入,分别建立主流烟气化学成分和感官得分的BP神经网络预测模型.隐含层节点为9,输入函数为Tansig,输出函数为Purelin.训练方法为梯度下降法.选择22个样本作为训练样本,其中19个作为测试样本,3个作为验证样本.训练的目标为允许误差0.000 1,最大迭代次数10 000次.预测结果与烟气常规化学检测和人员实际评吸结果比较,相对标准偏差小于5％,达到了较好的预测结果.该模型对于预测卷烟主流烟气成分的释放量和感官评价具有指导意义.
展开▼
