Short-circuit Fault Identification And Distance Measurement Of AT Traction Network Based On Deep Learning

The AT traction power supply system has been widely used in the power supply of highspeed railways in our country.The traction network as the core component of the traction power supply system.The scope of the fault will expand and cause serious impact when the fault of the traction network cannot be dealt with in a timely and effective manner.Identifying the type of failure and determining the location of the failure accurately and quickly is conducive to the repair of the traction network by the railway staff,it can reduce the time of failure and reduce the influence of the fault.In order to solve the problem that the huge demand for data in deep learning methods with the less data of transmission line failure on AT traction power supply system,this paper proposes a deep learning method based on the rapid identification and distance measurement of AT traction network short-circuit faults.The electromagnetic transient software ATP-EMTP is used to simulate the short-circuit fault on the transmission line of the AT traction power supply system to obtain the short-circuit data set of the traction network.First,the paper collected 4275 sets of experimental data for the production of deep learning data sets by using simulation software to complete the model building of the fully parallel AT traction power supply system and conducts traction network short-circuit failure experiments under different conditions.Through the combination of convolutional neural network and long-term neural network,a model suitable for short-circuit fault identification and ranging of AT traction network is constructed.The structure and parameters of the model are determined through experimental comparison,and the model is generated after optimization using the data set.The model can identify the short-circuit faults between the contact lines,feeders,and rails of the traction network,and complete the distance measurement of the fault distance.Through the verification experiment of the model,the model identification is accurate and the ranging accuracy is high.