摘要
针对国内外高速铁路的快速发展,道岔故障严重影响行车安全及运输效率,本文提出一种基于隐马尔科夫模型的道岔故障诊断方法,通过增加道岔设备的潜在故障状态,将道岔设备的状态进行多状态细分。采用基于Fisher准则函数和主成分分析的方法进行特征提取,矢量量化处理后,建立不同故障模式下的HMM模型,通过比较测试数据与训练得到的不同HMM模型的匹配值进行故障诊断。利用京广铁路长沙南某型号道岔连续动作功率数据,对模型的性能进行测试,完成了故障诊断的实现与验证。仿真结果表明,采用四维特征信息时,其训练时间相对于其他机器学习方法有了较大提高,正确率达到90%以上,且该方法将道岔状态进行细分,通过分析每种状态之间的状态转移,可以预测道岔故障,从而进行道岔设备健康状态监测。
With the rapid development of high speed railways, switch failure seriously affects the safety and transportation efficiency. This paper presented a Hidden Markov Model based method that aims to detect fail ure progression in turnout systems. The failure degradation states were subdivided through the increase of the potential failure state of the turnouts. The Fisher criterion function and principal component analysis （PCA） were used to extract and select the feature of the corresponding failure modes. The HMM models under differ ent failure modes were established after vector quantization processing to perform the fault diagnosis by compa ring the matched values of different HMM models obtained by the test data and the training. The performance of the model was tested by using the continuous operating power data of a certain type of a turnout in Changsha South Station of Beijing Guangzhou high speed railway to complete the realization and verification of fault diag nosis. The experiment results show that the training time relative to other machine learning methods has great ly improved when the four dimensional feature information is used, with accuracy of above 90%. Meanwhile, the method subdividing the failure degradation states can predict turnout faults based on the analysis of state transition between each state to improve the switch equipment health management.
作者
许庆阳
刘中田
赵会兵
XU Qingyang;LIU Zhongtian;ZHAO Huibing(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)
出处
《铁道学报》
EI
CAS
CSCD
北大核心
2018年第8期98-106,共9页
Journal of the China Railway Society
基金
中央高校基本科研业务费专项资金(2017JBM009)
