中国铁道科学2023，Vol.44Issue(6)：143-152,10.DOI:10.3969/j.issn.1001-4632.2023.06.15
基于均值电压和极限学习机的牵引逆变器开路故障诊断方法
Open-Circuit Fault Diagnosis Method for Traction Inverter Based on Average Voltage and Extreme Learning Machine
摘要
Abstract
Aiming at the characteristics of high speed and precise localization for open-circuit fault diagnosis of traction inverter in rail transit industry,an open-circuit fault diagnosis method for inverter based on average voltage and extreme learning machine is proposed.Firstly,the open-circuit fault of three-phase two-level topology inverter is analyzed,and the characteristics of open-circuit fault are summarized.Secondly,the average voltage is extracted from the fault characteristics as a basis for fault detection,and the fault feature vector is constructed using the related parameters of stator current.Finally,the ELM fault diagnosis model is trained offline,the fault classifier is generated and input into the online diagnosis process,to achieve the framework construction of the proposed fault diagnosis method.Robustness test is carried out based on normal working data,210 sets of fault data are obtained by setting different fault time,different fault types,different speed and load conditions,and then online diagnostic test is carried out,which is compared with the test results of mSVM,DT and RF for verification.The results show that compared with the other three methods,the proposed method has higher robustness under various normal conditions.As for the online diagnostic test,only 1 group of data is misdiagnosed,with the test accuracy reaching 99.5％.The training time is 0.28 s and the fault diagnosis time is 21 ms,which are both the shortest of these methods.The proposed method is suitable for applications requiring fast diagnosis,precise localization and strong robustness.关键词
牵引逆变器/开路故障/均值电压/极限学习机Key words
Traction inverter/Open-circuit fault/Average voltage/Extreme learning machine分类
交通工程引用本文复制引用
王清永,李文鹏..基于均值电压和极限学习机的牵引逆变器开路故障诊断方法[J].中国铁道科学,2023,44(6):143-152,10.基金项目
国家重点研发计划项目(2018YFB1201801-4) （2018YFB1201801-4）
