基于MEEMD香农熵-LSSVM的高速列车蛇行失稳诊断方法  

Hunting instability of high-speed train diagnose method based on modified ensemble empirical mode decomposition Shannon entropy and LSSVM

作　　者：叶运广 宁静[1] 种传杰 崔万里[1] 刘棋[1] 

机构地区：[1]西南交通大学机械工程学院,成都610031

出　　处：《计算机应用研究》2017年第4期1097-1100,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(51475387);中央高校基本业务费专项基金资助项目(2682014CX033);四川省科技创新苗子工程资助项目(2015102)

摘　　要：针对列车高速运行时易出现蛇行失稳这一问题,提出了一种改进的集合经验模态分解(modified ensemble empirical mode decomposition,MEEMD)香农熵—最小二乘法支持向量机(least squares support vector machine,LSSVM)的高速列车蛇行失稳诊断方法。首先通过MEEMD对列车330~350 km/h时转向架构架的横向振动信号进行分解,得到固有模态函数(intrinsic mode function,IMF),再通过Hilbert变换(HT)分析其时频聚集性,同时提取IMF分量的香农熵特征,最后用LSSVM进行训练和识别。结果表明,转向架蛇行失稳状态下的时频分布的聚集性较正常状态下好,并且MEEMD香农熵-LSSVM方法的识别率和计算耗时优于EEMD-SVM方法,识别率达到96.67%。Aiming at addressing the issue of hunting instability for the train at a high speed, this paper presented a new methodology which combined modified ensemble empirical mode decomposition （ MEEMD）, Shannon entropy features and least squares supported vector machine （LSSVM） to diagnose hunting motion state of high-speed train. This method used MEEMD to decompose the vibration signal under 330 - 350 km/h and got IMF. It used Hilbert transformation （HT） to analyze the spec- trum of hunting motion signal, and finally used LSSVM to train and recognize Shannon entropy features of IMFs. The results show that distribution of bogie frequency is highly centralized, and the methodology of MEEMD Shannon features-LSSVM can accurately recognize the unsteady state of hunting motion, up to 96.67%. Furthermore, the accuracy and calculation time are superior to ensemble empirical mode decomposition-support vector machine（EEMD-SVM）.

关 键 词：蛇行运动 改进的集合经验模态分解(MEEMD) Hilbert-Huang变换(HHT) 香农熵 最小二乘法支持向量机(LSSVM) 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]