摘要
基于支持向量机的传统模拟电路故障诊断方法对新故障无检测能力,且可扩展性较差。针对该问题,提出结合一类支持向量机(OCSVM)和多类支持向量机(MCSVM)的故障诊断方法。该方法采用OCSVM对故障数据进行检测和初步分类,采用MCSVM提高分类性能,以弥补OCSVM分类能力的不足。对OCSVM算法进行改进,以提高其检测和分类性能。通过模拟电路故障诊断实验验证OCSVM改进算法和联合故障诊断方法的有效性。
The traditional diagnosis method based on support vector machine lacks new type fault detection ability and expansibility. To solve this problem, a novel diagnosis method based on the combination of One-class SVM（OCSVM） and Multi-class SVM（MCSVM） is proposed. In this method, the OCSVM module is applied to the detection and preliminary classification of fault data, and the MCSVM module is used to improve the classification capability, which is a shortcoming of the OCSVM. Besides, the OCSVM algorithm is improved to perfect its detection and classification ability. Through an analog circuit fault diagnosis experiment, result shows the effectiveness of the improved OCSVM algorithm and the combined fault diagnosis method.
出处
《计算机工程》
CAS
CSCD
2012年第4期170-173,共4页
Computer Engineering
基金
国家部委基金资助项目
关键词
模拟电路故障诊断
支持向量机
一类支持向量机
决策函数
正负类间隔
参数选择
analog circuit fault diagnosis
Support Vector Machine（SVM）
One-class SVM（OCSVM）
decision function
distance between positive and negative class
parameter selection
