Study On Locomotive-Grid Electrical Matching Problem Of Sichuan-Tibet Railway And Its Monitoring Algorithm

In recent years,there have been many electrical matching problems between locomotive and traction power supply system(TPSS)(also known as “locomotive-grid electrical matching problem”),which have caused interference to the safety and reliability of electrified railway,such as high-order harmonic resonance,reactive power /network voltage oscillation,abnormal traction network voltage on long and steep slope section,low frequency oscillation and neutral section electromagnetic interference.Due to its special environment and great strategic significance,Sichuan-Tibet railway has higher requirements for the stability of electrical matching between locomotive and TPSS,the generation mechanism and monitoring algorithm of which should be studied.Firstly,the establishment of locomotive and TPSS model is the basic work.This paper establishes the harmonic current source model of locomotive,power source model of locomotive and TPSS unified chain model,and puts forward the circuit analysis methods,including power flow calculation(steady-state analysis)and transient analysis.Secondly,on the premise of considering the particularity of Sichuan-Tibet railway such as the direct power supply mode with negative forward line,bilateral power supply structure in some sections,long ramps and high-power locomotives of different speed grades,this paper analyzes the mechanism of the typical electrical matching problems,and determines the influencing factors based on locomotive and TPSS model.And based on measured data,the time-frequency characteristics of various problems are summarized.Thirdly,in order to further realize the data analysis,in this paper,time-frequency feature extraction tool—short time Fourier transform(STFT)and data mining tool—neural network are used synthetically,and their basic properties and parameter selection are analyzed.Finally,on the basis of data analysis,this paper proposes data preprocessing based on autoencoder,identification algorithm based on STFT and convolutional neural network(CNN),and identification algorithm based on long short-term memory(LSTM)network.The results show that different methods have their own advantages and disadvantages and application scenarios,so we should choose the appropriate algorithm from the actual needs.