Research On Fault Diagnosis For Jointless Track Circuit Based On Deep Brief Network

In recent years,high-speed railway has developed rapidly and becomes one of the popular means of transportation for people's daily travel.At the same time,it meets higher requirements for safety and operating efficiency.The track circuit is an important part of Chinese Train Control System,and it is an indispensable ground signal equipment to ensure the safe operation of the train.Once the track circuit fails,it will directly affect the train operation efficiency and even endanger the operation safety.Currently,the fault diagnosis of track circuit is still relying on maintenance personnel experience,which could have the several problems such as low fault detection efficiency,large amount of labor.A major problem faced by maintenance personnel is how to improve the intelligent level of track circuit fault diagnosis and make fault diagnosis efficiently.Based on the characteristics of the track circuit,this thesis,combing the existing research results,studies the track circuit diagnosis problem from the following steps.(1)Taking ZPW-2000 A jointless track circuit as the research object,the equipment configuration and working principle of the system are introduced,and 15 common failure modes are summarized.Under the normal state,the equivalent theoretical model of track circuit is established by using four-terminal network theory.According to track circuit actual parameter,each monitoring point value is deduced using the distributed parameter method.By setting different parameters to theoretical model,fault data collection is completed.(2)A fault diagnosis algorithm for jointless track circuit based on deep belief network model is proposed,and the overall block diagram of the diagnosis process is provided.Based on the dimensions of input samples and fault category,this thesis constructs the deep belief network fault diagnosis model.Meanwhile,unsupervised pre-training and supervised fine-turning are carried out by contrast divergence algorithm and back propagation algorithm,so as to mine the internal characteristics of data and establish the complex nonlinear relationship between characteristics and states.Considering the influence of the deep belief network structure on the fault diagnosis results,the thesis sets up multiple hidden layer combinations and uses theoretical data for simulation analysis.The simulation results show that the 6-layer structure fault diagnosis model has the best performance and can achieve better diagnostic results.(3)For the problem that deep brief network structure needs to be adjusted manually in training process,a parameter optimization framework based on dynamic addition algorithm and particle swarm optimization algorithm is proposed to get the optimal parameter configurations.First,the dynamic addition algorithm is used to compensate the number of layers through the error and the error decline rate to obtain the simplest network structure that meets the accuracy requirements.Then,under the premise of classification error rate as the fitness function,the particle swarm optimization is selected to optimize the number of hidden layer nodes,and the best deep brief network structure is determined to solve the problem of track circuit fault diagnosis.The simulation results indicate that the optimized deep belief network model can not only have high training efficiency but also achieve 98.864% fault identification rate.(4)Combined with the above research,a jointless track circuit fault diagnosis system is designed by using C# language,My SQL database technology and mixed programming technology.Through applying the system to diagnose actual jointless track circuit fault data,the feasibility and effectiveness of the system are verified,and those algorithms are proved feasible practical value in actual system.