基于信息几何的卷烟焦油SVM(支持相量机)预测  

Cigarette tar delivery prediction by SVM based on geometric information

作　　者：王德吉[1,2] 李广才[1] 栗卫军[1] 

机构地区：[1]中国烟草总公司职工技术培训中心培训处,郑州市北林路7号450008 [2]中科院合肥智能机械研究所,合肥市科学岛1130信箱230031

出　　处：《中国烟草学报》2009年第4期22-25,共4页Acta Tabacaria Sinica

基　　金：国家自然科学基金资助项目(No.60075022)

摘　　要：由于样本少且是非线性关系,卷烟配方中焦油量预测非常困难,为此引入能较好的解决小样本非线性预测问题的SVM(支持向量机)进行卷烟焦油预测,其中的核函数选择是关系到预测精度的关键,而现有方法仅能试验试凑。本文从信息几何的角度,通过保角映射,给出核函数构造的一般方法,提高了预测精度和效率并通过试验证明该模型能更准确的预测卷烟焦油量。Cigarette tar delivery is difficult to predict because of inadequate samples and nonlinear relationship among variables. Hence a SVM （Support Vector Machine） forecast method based on information geometry was proposed. In SVM based method, kernel function is very important to the prediction accuracy. A new method was given by conformal mapping from the perspective of geometric information, which can enhance accuracy and efficiency. Experiment was carried out to test the performance of the presented method. Result showed that the SVM based on information geometry can predict tar delivery better than currently used method.

关 键 词：SVM(支持向量机) 卷烟焦油量 信息几何 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]