摘要
针对粒子群算法对支持向量机参数进行优化时存在的收敛速度慢、分类准确率不高的问题,通过引入Fisher准则评估每个特征向量粒子的适应度得到最优特征子集,提出了一种基于Fisher准则下粒子群算法优化支持向量机(FIPSO-SVM)的新分类方法,该方法的目标是尽可能地加大类间间隔和减小类内间隔。采用滚动轴承数据集在时域和频域上得到32组特征向量,测试该方法在4种工作状态下的分类效果,最后,使用不同核函数和2种不同算法将全样本特征向量与最优特征向量子集的SVM分类结果进行对比。结果表明,FIPSO-SVM分类器不仅能够识别故障产生的位置,还能区别故障损伤的程度,FIPSO-SVM分类器具有更高的分类精度和更快的收敛速度,值得进一步在工程领域内推广。
In order to solve the problem of slow convergence rate and low classification accuracy for the optimization of SVM parameters in particle swarm optimization algorithm,this paper proposes a new classification method that uses the improved particle swarm optimization（IPSO）based on the Fisher criterion to optimize the parameters of SVM（FIPSO-SVM）.In this method,the Fisher criterion is applied to assess the fitness of each feature vector particle in order to get the best feature vector subset.The purpose of this method is to increase the between-class intervals and reduce the within-class intervals as much as possible.The effectiveness of the method was tested by using the rolling bearing data set to obtain 32 sets of eigenvector in the time domain and frequency domain in 4 working conditions,and the whole sample eigenvectors and the SVM classification results were compared by using different kernel functions and two different algorithms.The results show that the FIPSO-SVM classifier can not only identify the location of the fault,but also distinguish the degree of fault damage.It has higher classification accuracy and faster convergence speed which is worthy of further promotion in the engineering field.
作者
吕明珠
苏晓明
陈长征
刘世勋
LYU Mingzhu;SU xiaoming;OHEN Ohangzheng;LIU Shixun(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China;Department of Automatic Control,Liaoning Equipment Manufacturing Professional Technology Institute,Shenyang 110161,China;CQC（Shenyang）North Laboratory,Shenyang 110164,China)
出处
《机械与电子》
2018年第7期49-54,共6页
Machinery & Electronics
基金
国家自然科学基金资助项目(51675350)
高校重点课题(2018XB01-4)
关键词
FISHER准则
粒子群算法
支持向量机
滚动轴承
故障诊断
Fisher criterion
particle swarm optimization
support vector machine
rolling bearing
fault diagnosis
