Research On The Fault Diagnosis Of Jointless Track Circuit Based On SOM Neural Network

In recent years,the high-speed railway construction of China is being carried out at full speed,as a result,train operation efficiency and safety of the train are getting more and more attention.Jointless track circuit is widely used in Chinese railway signal systems.Tuning area,one of the important components of jointless track circuit,is set between adjacent track circuits.It has taken well targeted measures to keep the transmission range of the signal within limit and avoid signal transmissions between different track circuits basing on the different resonance characteristics for different track circuit signals.Fault of tuning area will directly influence the train control system,which results in low efficiency of operation and safety of the train.Therefore,it is of great importance to extract fault signals of tuning area.Firstly,four terminal network model of normal state and tuning area fault state of insulated track circuit was established under shunting condition.jointless track circuit under the condition of shunt state was established.Using distributed parameter method to derive corresponding model parameters.According to the four terminal network theory,the four terminal network model of different state of jointless track circuit under the condition of shunt state is established.Based on this,the shunt current amplitude envelope functions expression is obtained.Then,setting simulation condictions based on the model.The shunt current amplitude envelope under the situations of jointless track circuit normal,tuning area and other factors fault can be simulated.The change law of the corresponding was obtained by analyzing the influence of different factors on the envelope amplitude of the shunt current curve.The shunt curr-ent amplitude envelope is decomposed respectively by Empirical Mode Decomposition(EMD),Ensemble Empirical Mode Decomposition(EEMD)and Complementary Ensem-ble Empirical Mode Decompostion(CEEMD),After this,the fault feature vector was extracted.Results shows that the CEEMD method suppresses the mode mixing and residual noise induced by EMD and EEMD.At same time it improve the calculation efficiency and effectively extracted the fault characteristics in the tuning area of jointless track circuits.Finally,based on the regulations of tuning area and shunt current amplitude envelope,the tuning area fault diagnosis of jointless track circuit with the Self-Organizing feature Map(SOM),neural network was proposed.In this thesis SOM neural network was constructed at first.A part of the feature vector was used to train the SOM neural network while the others were used as a test on SOM neural network.Based on the above,tuning area signle fault and mix fault would be recognized and classified,the diagnosis can be finished.The method which was raised in this thesis is validated by the actual data.On the onehand,the accuracy of shunt current amplitude envelope model for jointless track circuit is verified.On the other hand,fault from the tuning area of jointless track circuit can be diagnosed effectively.Compare with the improved BP neural network,the result of simulation shows that using SOM neural network is more efficient and accurate in tuning area fault diagnosis.