神经网络与模糊C均值在烟料替换中的应用

Application of Neural Network and FCM in the Substitution of Tobacco Raw Material

针对各烟草企业在叶组配方设计时仍使用国家分类标准,结合专家经验来选择烟叶原料,此方法未考虑烟叶内在质量及理化特性。提出了一种应用神经网络技术与模糊数学相结合的烟叶原料分类替换方案。对烟叶样本集的具体理化特性加权后,采用Support Vector Machines(SVM)神经网络将样本集分成专家经验集和未知分类集,再对两类样本集各自采用Self-Organizing Feature Map(SOM)神经网络聚类。对于聚类后存在的三样本集问题,采用Fuzzy Center Means(FCM模糊C均值)法找出其最佳替代原料。结果表明此方法比传统分类方法具有更好的灵活性与适应性,是一种研究烟叶原料替换的有效方法。

Aiming at each tobacco enterprise still adopting national classification standard combining with expert experience to choose tobacco raw material in the design of tobacco compound formula,this approach doesn＇t consider the inherent quality and the physicochemical characteristic of tobacco. Put forward the tobacco raw material classification substitution scheme which integrates neural network technology with fuzzy mathematics. Specific physicochemical characteristic of tobacco specimen collection multiplies by weight,employing SVM neural network dividing the specimen collection into expert experience collection and the unknown classified collection,then applying SOM neural network clustering the two kinds of specimen collections respectively,aiming to existing clustering-after tri-specimen collection issues,using the FCM method to discover its best substituted raw material. The result showing that this method is more flexible and adaptable than traditional categorization method,it is an effectual approach in the study of tobacco raw material substitution.