烟叶致香成分相关性分析研究

Correlation Analysis On Aroma Constituents of Tobacco Leaves

烟叶致香成分数据具有"小样本、高维数、模糊和非线性"等特点,传统的统计分析方法难以有效解决其相关性问题。本文综合运用M5P模型树、mRMR特征选择以及神经网络等多种智能方法解决致香成分之间复杂的非线性问题,并结合行业专家经验,客观地对多种方法结论进行验证和评价,得出具有概括性的最终特征参数。综合分析方法拟补了传统分析方法的不足,提高了数据分析结果的准确度。

Because aroma constituents of flue-cured tobacco leaves are characteristic of small sample,high-dimension,fuzzy and nonlinearity,Its relevancy is difficult to be resolved by adopting traditional statistics methods.In this article,several computational intelligent approaches including M5P model tree、mRMR feature extraction and neural network were comprehensively adopted to resolve nonlinear problem between different aroma constituents.In addition,combining abundant expert experiences,conclusions given by a variety of methods were chosen and final characteristic parameters were obtained.Comprehensive analysis methods make up the inadequacy of traditional analysis methods and improve the accuracy of the data analysis results.