Research On Fault Diagnosis Of Traction Power Supply System Based On Data Drive

With the steady improvement of China’s economic strength and technological strength,high-speed railway,as an important part of modern transportation,is developing vigorously in the direction of higher speed and more intelligent.Traction power supply system as the core of the whole high-speed railway,its reliable and stable operation,will directly affect the running state of high-speed train.Due to longterm service in complex and harsh external environment,traction power supply system will inevitably have some failures.When the traction power supply system breaks down,how to quickly find out the cause of the failure and restore the power supply is of great significance to maintain the normal railway order.When the traction power supply system fails,the comprehensive automation system of traction substation will record a large number of fault recording data.The utilization rate of these recording data is low,but it contains a large number of fault information to be mined and sorted out.Aiming at the problem of fault diagnosis of traction power supply system,this paper will start with these historical fault recording data,extract characteristic quantities that can represent different fault types,and establish fault classification model to achieve the purpose of fault diagnosis.In this paper,from the one AT the traction substation integrated automation system has collected from 2010 to 2021,all of the samples of fault wave record,according to the frequency of failure,the foreign invasion,lightning fault,the sound,because of the locomotive itself causes trip and over load fault as the research object of this article,and analyses the cause of such failure.Through the analysis of recorded wave data,it is found that when the traction power supply system fails,its feeder voltage will produce a large number of nonlinear and non-stationary signals,and these signals contain a large number of fault information,so the feeder voltage data at the time of failure is selected to extract the fault characteristic quantity.In this paper,the ensemble empirical mode decomposition(EEMD)algorithm is used to decompose the feeder voltage when the traction power supply system fails.Intrinsic mode function(IMF)component is used as the eigenmatrix and singular value decomposition is carried out.Then singular value entropy is obtained by solving the singular value entropy theory.Meanwhile,the fuzzy entropy of all IMF components is calculated according to the fuzzy entropy theory.By comparison,it is found that the singular value entropy and fuzzy entropy obtained by solving the first three IMF components have great differentiation in the five fault states and can be used as characteristic quantities to characterize different fault types.In order to diagnose the faults of traction power supply system,this paper established a multiclassification support vector machine(SVM)model optimized based on genetic algorithm.The extracted singular value entropy and fuzzy entropy were input into the SVM model as feature quantities for training and testing.The final results show that the accuracy of this method is 96%.It can effectively diagnose the five kinds of traction power supply system faults studied in this paper.Finally,the human-computer interaction software of traction power supply system is developed by means of mixed programming of MATLAB and C++,which mainly includes failure import module,data analysis module and failure alarm module.The software can analyze and self-learn the fault recording data when the traction power supply system fails,and provide the basis for the staff to deal with the fault.