摘要
针对局部切空间排列算法面临的无法利用样本标签信息和不能高效处理增量式维数约简问题,提出一种新的增量式监督局部切空间排列算法(Incremental Supervised Local Tangent Space Alignment,ISLTSA)。为充分利用训练样本标签信息,在LTSA算法的基础上加入散度矩阵,构造新的最小目标函数,使得高维样本的低维嵌入坐标同类聚集、异类分离。对于新增样本可能影响部分训练样本局部邻域,更新全局坐标矩阵,获取训练样本低维坐标和新增样本低维坐标,并作为初值进行特征值迭代实现所有样本全局坐标的更新。结合支持向量机分类算法,将ISLTSA算法应用于齿轮箱的故障状态识别,实验分析验证了该方法的监督学习能力,可提高故障状态识别率,并具备增量学习能力,可降低维数约简方法的复杂度。
Aiming at that the local tangent space alignment( LTSA) algorithm could not use samples 'label information and could not fast process incremental dimension reduction problems,a new incremental supervised local tangent space alignment( ISLTSA) algorithm was proposed.To make full use of the label information of training samples,the divergence matrix was added into the LTSA algorithm to construct a new minimum objective function.The lower dimensions were made to embed coordinates for homogeneous clustering and heterogeneous separating.The incremental samples might affect the local neighborhood of partial training samples.Then the global coordinate matrix was updated to get the lower dimension coordinates of both training samples and the incremental ones,the lower dimension coordinates were taken as initial values to do eigenvalue iteration and realize updating the global coordinates of all samples.Combined with the classification algorithm of support vector machine,the proposed ISLTSA algorithm was applied in gearbox fault diagnosis.The tests verified the supervisory and learning capacity of the proposed method,it was shown that the new method can improve the fault recognition rate; it has an incremental learning ability,and can reduce the complexity of the dimension reduction method.
作者
佘博
田福庆
梁伟阁
汤健
SHE Bo;TIAN Fuqing;LIANG Weige;TANG Jian(Department of Weaponry Engineering, Naval University of Engineering, Wuhan 430000, China;Faculty of hfformation Science, Beijing University of Technology, Beijing 100124, China)
出处
《振动与冲击》
EI
CSCD
北大核心
2018年第13期105-110,129,共7页
Journal of Vibration and Shock
基金
国家自然科学基金(61573364
61640308)
关键词
增量式学习
监督局部切空间排列
故障诊断
支持向量机
incremental learning
supervised local tangent space alignment
fault diagnosis
support vector machine