DAG SVM的结构优化研究及其在故障诊断中的应用
Support vector machine with structure optimized decision directed acyclic graph and its application to fault diagnosis
作者：陈思羽(四川大学电子信息学院)；宁芊(四川大学电子信息学院)；周新志(四川大学电子信息学院)；赵成萍(四川大学电子信息学院)；雷印杰(四川大学电子信息学院)
Author：CHEN Si-Yu(College of Electronics and Information Engineering, Sichuan University)；NING Qian(College of Electronics and Information Engineering, Sichuan University)；ZHOU Xin-Zhi(College of Electronics and Information Engineering, Sichuan University)；ZHAO Cheng-Ping(College of Electronics and Information Engineering, Sichuan University)；LEI Yin-Jie(College of Electronics and Information Engineering, Sichuan University)
收稿日期：2014-04-16 年卷（期）页码：2015,52(2):299-305
期刊名称：四川大学学报: 自然科学版
Journal Name：Journal of Sichuan University (Natural Science Edition)
关键字：支持向量机; 有向无环图; 多分类; 故障诊断
Key words：Support vector machine; Directed acyclic graph; Multi class classification; Fault diagnosis
基金项目：国家973计划项目（2013CB328903 2）
中文摘要
有向无环图支持向量机(DAG SVM)是一种新颖且使用广泛的多分类算法. 传统DAG SVM由于需要训练的SVM分类器较多, 在工程中训练耗时长. 又由于传统DAG SVM分类效果受到结构排序影响, 导致其分类效果具有随机性. 针对以上两个问题, 通过结构重组减少SVM分类器个数从而缩短了训练时间, 通过对训练数据的重新划分计算产生了最优分类排序, 提高了分类正确率. 仿真测试与工程实践证明, 本文方法相对传统DAG SVM方式, 能缩短训练时间, 且拥有更高的分类正确率.
英文摘要
Directed Acyclic Graph Support Vector Machine(DAG SVM)is one of the variants of SVM, which has been used in a wide range of fields. However, original DAG SVM requires lots of single SVM classifiers as well as training effort. Moreover, the classification results have randomness due to the DAG’s different structure sequencing. In order to solve the mentioned two problems above, the proposed method reduces the number of SVM classifiers by restructuring thereby enhance the training efficiency It can also improve the rate of correct classification through the training data division and calculation. The simulation tests and the engineering practices prove that the structure optimized DAG SVM investigated in this paper is able to improve the training efficiency and obtain a higher classification accuracy
【关闭】
