Study On Correlation Analysis Of Statistical Failure Data In High-speed Railway Traction System

The reliability of traction power supply system is a prerequisite for the safe and punctual operation of high-speed rail lines.It can provide recommendations for avoiding accidents through the analysis of rail traction fault severity and associated factors as well as fault diagnosis of traction transformers.Due to the limitation of fault-data acquisition in high-speed traction power system,there exists the different capacities of data samples,which is required to establish two analytical models for analysis.In the context of large-sample data,statistical failure data of high-speed traction should be analyzed,in order to minimize high-speed traction power failure times and reduce the severity of the fault.First,fault severity of traction is appointed to the dependent variable,and candidate independent variables ought to be selected from variables among traction internal structure's and external environmental factors,while backward stepwise selection method is used to find the correlation between the candidate independent variables and dependent variables.Second,the correlation-analysis model,which is suitable for analysis of large data,is built between rail traction fault severity and influence factors through using cumulative Logistic Model.Finally,the model is verified from three sides including proportional test,goodness-fit test and forecast-accuracy test.Statistical failure data of power supply system in recent years are analyzed,and the results show that there exists a significant correlation between traction fault severity and variables among the service life of the system,maintenance times,lightning strikes and wind speed.In the context of small sample data,the traditional model of grey relational analysis is improved.First,the definition of generalized coefficient is put forward based on the coefficient variation,which is used as a new standard to estimate grey-size correlation and reconstruct the correlation-calculation model.This model proceeded from overall,takes the relevance of the positive and negative aspects into account and avoids problems caused by point-calculation and the weighted average correlation offset.Due to the volume fraction of dissolved gases in transformer oil in stable operating state,a fixed reference sequence can be established.In this paper,an improved model of grey relational analysis is added to failure diagnosis of traction transformers,which takes advantage of the grey-correlation analysis of small data and the data characteristics of dissolved gas in oil,which reduces system complexity of the error-analysis and improves the diagnostic accuracy of the system.Finally,the established correlation analysis model in both cases is verified in the case study.