一种SVM-DS决策融合方法及在高铁故障中的应用  

A decision fusion method of SVM-DS and its application in high-speed rail fault

作　　者：朱菲 金炜东[1] Zhu Fei;Jing Weidong(Department of Electrical Engineering,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南交通大学电气工程学院,四川成都610031

出　　处：《电子技术应用》2018年第7期127-130,134,共5页Application of Electronic Technique

基　　金：国家自然科学重点基金项目(61134002)

摘　　要：针对信息融合分类中DS理论基本概率赋值函数(BPA)一直难以解决的问题,提出了一种基于SVM和DS理论的决策融合方法。利用Platt概率模型将不同核函数SVM分类器的硬输出转化为概率输出,并将混淆矩阵作为计算各分类器局部可信度的依据。根据SVM的后验概率和分类器的局部可信度来建立基本概率赋值函数,再通过DS融合做出最终决策。将该方法应用于高铁故障数据,实验结果表明,该构造BPA的方法在实际问题中有效且合理,该决策融合方法与单一分类器相比,能稳定地提高分类准确率。In order to solve the difficulty of obtaining the Basic Probability Assignment（ BPA） of DS evidence theory in the practical application, a decision fusion method combing SVM and DS evidence theory is proposed. The probability model of Platt is used to transform the hard outputs of different SVM classifiers into probability outputs, and the confusion matrices are used to calculate the local credibility of classifiers. The BPA function is established based on the posterior probability of SVM and local credibility of classifiers. Finally, the final decision is made through DS fusion based on BPA. Applying this method to the high-speed rail fail-ure data, the experimental results show that the method of constructing BPA is effective and reasonable in practical problems, and the method of decision fusion can steadily improve the classification accuracy compared with the single classifier.

关 键 词：支持向量基 DS理论 决策融合 局部可信度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] U279[自动化与计算机技术—计算机科学与技术]