Prediction of Flue-Cured Tobacco Leaves Aroma Quality and Volume Based on genetic algorithm and support vector machine

To reveal the relationship between the aroma quality of roasted tobacco and other quality indicators, the conventional chemical composition and physical characteristics of tobacco samples (central leaves) from 22 tobacco producing provinces (cities and autonomous regions) in China were used to integrate the data. The genetic algorithm and support vector machine regression in artificial intelligence algorithms were used for screening variables and establishing prediction models, and the model was evaluated by using the modeling, leave-one-out and prediction. The results showed that the chemical composition indexes that have a greater influence on tobacco aroma quality are nicotine, total nitrogen and reducing sugar, and the physical properties are thickness and equilibrium water content. The chemical composition indexes that have a greater influence on tobacco aroma volume are nicotine and total nitrogen, and the physical property indexes are tensile strength, stem rate and equilibrium water content. The prediction results of the support vector machine regression model for tobacco flavor are less than 0.5% of the measured values. The proportion of samples with absolute errors within 0.5 ranged from 89.32% to 98.24%, and the absolute errors ranged from 0.17 to 0.26; the proportion of samples with absolute errors within 0.5 ranged from 84.94
^%
to 97.61%, and the absolute errors ranged from 0.16 to 0.27. In summary, the fusion data of standard chemical composition and physical characteristic indexes indicate that a tobacco flavor quality forecasting model with high accuracy can be established through the joint application of artificial intelligence algorithms.