支持向量机动态多分类方法  

The Dynamic Muliti-classification Method of Support Vector Machine

作　　者：房汉鸣 税爱社[1] 汪辉 宗福兴[3] Fang Han-ming;Shui Ai-she;Wang Hui;Zong Fu-xing(Dept. of Logistics Information & Logistics Engineering Chongqing 401331,China;Dept. of Engineering Equipment Management and Guarantee,Institute of Engineer Corps,Xuzhou Jiangsu 221004,China;Dept. of Management Science & Engineering,LEU,Chongqing 401331,China)

机构地区：[1]后勤工程学院后勤信息与军事物流工程系,重庆401331 [2]工程兵学院工程装备管理与保障系,江苏徐州221004 [3]后勤工程学院管理科学与工程系,重庆401331

出　　处：《后勤工程学院学报》2017年第2期90-96,共7页Journal of Logistical Engineering University

摘　　要：针对支持向量机分类中存在盲区的问题,在分析多分类支持向量机构建方法的基础上,采用二叉树支持向量机,通过类型知识库动态更新,建立了二叉树支持向量机动态多分类模型;选取径向基函数作为核函数,给出了提高多分类模型性能的核参数和惩罚因子寻优遗传算法;以卫星整流罩空调系统表冷器控制单元为仿真实验对象,建立了传感器偏置故障、漂移故障、完全故障和精度下降故障的多分类模型,验证了提出方法的有效性。To eliminate the classification blind area in the support vector machine,the binary tree-support vector machinewas adopted on the basis of analyzing the construction method of the multi?classification support vector machine.The dynamic multiclassification model of binary treesupport vector machine was established by updating the category knowledge base.The radial ba?sis function was preferred as the kernel function.The optimization genetic algorithm of nuclear parameter and the punishment facto rwere put forward to improve the multi-classification classification model.The surface cooler in fairing air conditioning system was taken as the simulation object.The multi-classification model of the sensor with the faults such as bias fault,drifting,complete failure and the decline in accuracy were established.The result shows that the effectiveness of the proposed method is verified well.

关 键 词：支持向量机 二叉树 动态分类 传感器故障 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]