The Establishment of Prediction Model of Inventory Tobacco Flavor Based on RBF Neural Network

In order to establish the prediction model of inventory tobacco flavor, the authors analyze the samples of 2009-2011 inventory tobacco in China Tobacco Chuanyu Industrial Co., Ltd. by using the RBF neural network method. The results showed that there was difference of the content of chemical components between different tobacco flavors, sugar content in clean aroma type tobacco was significantly higher than the others, and chlorine content in clean aroma type tobacco was much lower than that of full-bodied type. The authors used principal component analysis to eliminate the chemical indicator collinear problem, and established prediction models based on RBF neural network of inventory tobacco flavor. The accuracy rate of the models was up to 90%. The sensitivity test showed that the clean aroma type tobacco model had the best sensitivity, the moderate type showed a lower sensitivity. Tobacco flavor can be predicted based on chemical components using the RBF neural network.