基于LMD形态滤波的LSSVM方法研究
来源:用户上传
作者:孟良,许同乐,马金英,蔡道勇
摘要:在轴承的故障诊断中,为了解决核函数在最小二乘支持向量机中参数选择困难及稀疏性差的问题,提出了局部均值分解(LMD)形态滤波的最小二乘支持向量机(LSSVM)方法。该方法首先利用LMD对信号进行分解得到PF分量,并对信号做相关分析去除虚假分量,形态滤波降噪后再进行LMD分解得到新PF分量,提取能量特征;其次,对LSSVM的核函数进行改进,解决核参数选择的问题;应用特征加权法对拉格朗日参数进行特征加权,取其加权平均值作为剪枝方法的阈值,降低稀疏性;最后将能量特征信号输入LSSVM中,对信息进行训练预测。实验表明,应用该方法能快速有效地对轴承故障信号进行自适应的分类及轴承故障的判断。
关键词:局部均值分解; 形态滤波; 剪枝方法; 最小二乘支持向量机; 故障诊断
DOI:10.15938/j.jhust.2022.01.012
中图分类号: TH707 文献标志码: A 文章编号: 1007-2683(2022)01-0092-08
There Search of LSSVM Based on LMD Morphology Filter
MENG Liang1,XU Tongle1,MA Jinying2,CAI Daoyong3
(1.School Mechanical Engineering, Shandong University of Technology, Zibo 255049, China;
2.School of Agriculture Engineering and Food Science, Shandong University of Technology, Zibo 255049, China;
3.Shandong Keda M&E Technology Co.,Ltd., Jining 272000, China)
Abstract:In the diagnosis of bearing, the LSSVM method research with LMD morphological filtering was put out in order to solve the problem about the kernel function parameter selection and the bad sparsity of least squares vector machine (LSSVM). First, the LMD was used to decompose the measured signal and PF components were obtained. The correlation analysis was carried out to remove the false components, and the noise of PF components was reduced by morphological filtering. The LMD decomposed the recombinational signal and obtained new PF components, and energy characteristics were got from the new PF component. Secondly, the kernel function of LSSVM is improved to solve the problem of kernel parameter selection. Lagrange parameters were weighted by feature weighting method, and their weighted average value was taken as the threshold of pruning method to reduce the sparsity. Finally, energy characteristics were put into LSSVM to train and predict. Experiments showed that this new method could fulfil adaptive classification of bearing fault signals and definite fault conclusion quickly and effectively.
Keywords:local mean decomposition; morphological filtering; pruning method; least squares support vector machine; fault diagnosis
0前言
Suykens等[1]于1999年提出了最小二乘支持向量C(least squares support vector machine, LSSVM)方法。将最小二乘线性系统引入到传统支持向量机(support vector machine, SVM)中得到LSSVM方法,传统SVM中的约束条件为不等式约束,LSSVM方法中则转换成等式约束条件,其训练过程就变成了对线性方程组的求解,提高了传统SVM的求解效率,降低了学习难度,性能在很大程度上得到改进[2-3]。但是,LSSVM丧失SVM稀疏性的同时仍存在核参数选择困难问题,成为制约LSSVM应用的障碍[4-5]。
nlc202205121015
转载注明来源:https://www.xzbu.com/8/view-15431080.htm