基于CEEMDAN与KFCM聚类的转辙机退化状态识别方法  

Degradation State Recognition Method of Switch Machine Based on CEEMDAN and KFCM Clustering

作　　者：张友鹏[1] 张迪 杨妮[1] 魏智健 ZHANG Youpeng;ZHANG Di;YANG Ni;WEI Zhijian(School of Automation and Electrical Engineering,Lanzhou Jiaotong University,Lanzhou Gansu 730070,China;Wuhan Ditie Operation Co.,Ltd.,Wuhan Metro Group Co.,Ltd.,Wuhan Hubei 430070,China)

机构地区：[1]兰州交通大学自动化与电气工程学院,甘肃兰州730070 [2]武汉地铁集团有限公司武汉地铁运营有限公司,湖北武汉430070

出　　处：《中国铁道科学》2023年第1期194-201,共8页China Railway Science

基　　金：国家自然科学基金资助项目(51967010)。

摘　　要：针对道岔转换设备在使用寿命内的功率信号特征提取与退化状态识别问题,提出基于自适应白噪声完备经验模态分解(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise,CEEMDAN)与核模糊C均值聚类(Kernel-based Fuzzy C-Means clustering,KFCM)相结合的转辙机退化状态识别方法。首先,对S700K转辙机采集的功率曲线数据进行模态分解,得到多个固有模态函数(Intrinsic Mode Functions,IMFs),通过IMFs的能量幅值获得表征数据退化过程的特征向量;然后,由KFCM算法对特征向量进行转辙机退化状态识别,并进行状态划分;最后,通过计算分类系数和平均模糊熵对该方法的分类性能进行综合评估,并与模糊C均值聚类(Fuzzy C-Means clustering,FCM)和GK(Gustafson Keseel)聚类算法进行比较。结果表明:该方法聚类效果准确率达95.6%,优于FCM和GK聚类算法,能对转辙机的退化状态进行科学划分,为铁路现场道岔设备健康状态监测提供理论支撑。In order to solve the problems of power signal feature extraction and of degradation state recognition of turnout switching equipment within service life,a degradation state recognition method based on the combination of Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and Kernel-based Fuzzy C-Means(KFCM)clustering is proposed.Firstly,the power curve data collected by the S700K switch machine is decomposed into several Intrinsic Mode Functions(IMFs)through modal decomposition;and the feature vectors representing the data degradation process are obtained through the energy amplitude of IMFs.Then,KFCM algorithm is used to identify the degradation state of switch machine for its feature vectors and the state division is conducted.Finally,the classification performance of the proposed method is comprehensively evaluated by calculating the classification coefficients and the average fuzzy entropy,and is compared with Fuzzy C-Means(FCM)clustering and Gustafson Keseel(GK)clustering algorithms.The results show that the clustering accuracy of this method reaches 95.6%,which outdoes FCM and GK clustering algorithms,and can scientifically classify the degradation state of switch machine and provide theoretical support for the health status monitoring of on-site railway turnout equipment.

关 键 词：退化状态 S700K转辙机 特征提取 KFCM聚类 聚类识别 

分 类 号：U284.72[交通运输工程—交通信息工程及控制]