Research On Fault Prediction For ZPW-2000A Railway Jointless Track Circuits

Track circuit is an important equipment of railway signaling system which plays an important role in ensuring the train operation safely and effectively. The major accidents may happen once the track circuit has failures. At present, the repair and maintenance standard of china’s railway takes "TBM" way which has low maintenance and troubleshooting efficiency. It cannot repair and troubleshoot in certain key protection zones and failure-prone sectors in advance. Wherefore, the intelligent failure prediction realization of track circuit equipment that can make the work of track circuit more secure and reliable is great significance.In this thesis, the ZPW-2000A railway jointless track circuit is used as the study object which is commonly used in China. Historical track circuit data collected by the centralized monitoring system, track circuit test data on-site and the knowledge of maintain experts are fully utilized to fault prediction for interval track circuit. In this thesis, the specific work have done as follows:Firstly, the principles of its work are analyzed according to the characteristics of ZPW-2000A railway track circuit. The common causes of failure and fault phenomenon are summarized and examined.Secondly, the gray prediction model based on mathematics and the SVR(Support Vector Regression) model are selected to predict and model. On the basis of test data on the track circuit of railway administration, establish track circuit fault prediction system based on GM (1,1) model and SVR model which use the sparse data. The feasibility and effectiveness of the model is verified by test data on the track circuit of railway administration. Through fault examples, the forecasting results of fault analysis are obtained. The approximate time that faults occur can be predicted based on test data, meanwhile it can be prepared for the post-forecast which uses the dense data.Thirdly, Because of poor convergence performance of traditional BP neural network and it is easy to fall into local minima. That is why the wavelet network track circuit fault prediction model is built by combining the wavelet transform and neural networks model.The wavelet neural networks can be validated by simulating based on the data of ZPW-2000 track circuit monitoring subsystem. By the comparison of predict consequence between BP network and wavelet network, that the wavelet network model is better than the BP network model is been confirmed. Through the fault case analysis, feasibility and practicality of the model can be verified.Finally, this thesis summarizes the work has done and prospect the further work in the future.