Research On Fault Diagnosis Method Of 25Hz Phase Sensitive Track Circuit Based On LSTM

Track circuit is one of the three major outdoor equipment of railway signal,which is of great significance for the normal operation and efficient transportation of railway system.However,because the track circuit works continuously for a long time and the working environment is extremely complex and bad,frequent faults occur on the railway site,which will affect the operation efficiency if it is light,and the safety of driving is endangered if it is serious.Therefore,it is very meaningful to design an efficient and intelligent method to assist the field staff to quickly and accurately determine the type of failure and take timely measures to minimize the loss.The main contents of this paper are as follows:(1)This paper takes 25 Hz phase sensitive track circuit as the research object,its equipment composition and working principle are introduced in detail,analyzes the specific factors leading to its red tape fault and poor shunting,and establishes its equivalent four terminal network model.The critical value of voltage is calculated to assist the staff to judge the working state of track circuit.(2)In this paper,combined with the project of track circuit monitoring and self diagnosis system,after learning the knowledge of artificial neural network,LSTM neural network is used as the fault diagnosis model.After analysis,three voltage parameters closely related to the working state of track circuit are selected as the input of LSTM fault diagnosis model,and take 4 kinds of common faults and normal conditions in the field,5 kinds of categories in total as the output of the fault diagnosis model.LSTM can remember the past information and be good at processing long sequence data,so it can not only be used to analyze and process fault data,but also be used to obtain the time sequence information contained in the sequence data of the same parameter,so as to obtain more fault features.(3)BP fault diagnosis model and LSTM fault diagnosis model are established by MATLAB programming.After data normalization,the two models are trained by training data,and the parameters are adjusted continuously,so that the model can fully learn the real mapping relationship between input and output.Then the test sample set is used to verify the diagnosis effect of the trained model,and the simulation results of the two fault diagnosis models are compared.The results show that the LSTM model has better classification effect on the five cases,the fitting degree between the actual and the expected output curve is higher,average RMSE is 0.0308 and the fault diagnosis accuracy rate is 93.2%.It is proved that the method in this paper is efficient and feasible.(4)The PSO algorithm is selected to optimize the network parameters of LSTM model.The simulation results show that the overall effect of PSO LSTM model is further improved,average RMSE reaches 0.0197,and the fault diagnosis accuracy rate is 94.6%.The results show that when the number of hidden layer neurons of LSTM is 15,the learning rate is 0.13,and the maximum number of iterations is 306,the diagnosis effect is better.The simulation results show that the proposed method can accurately diagnose the common faults of track circuit,and the diagnosis performance is good.Compared with BP neural network,the diagnosis accuracy and precision of this method are higher.Moreover,the diagnostic accuracy of the optimized LSTM model is improved by 1.4%,and the error is reduced by 0.0111.40 pictures,13 tables,38 references.