Abstract：The electrified railways system is a complex system of traction power supply system and train

coupling. In order to ensure the safe operation of trains under complex working conditions，it is necessary to

identify the disturbance of the vehicle-network coupling system in a timely and accurate manner. To solve this

problem，a temporal convolutional neural network（TCN）was used to identify the disturbance of the vehiclenetwork coupling system. Firstly，the characteristics of each disturbance type of the system were analyzed，and a TCN network structure suitable for extracting time-domain features was built according to the analysis，the voltage at the main circuit breaker of the train was sampled with 1 s time window of，and the down-sampling operation was carried out to reduce the data size to reduce the number of network parameters，reduce the amount of calculation，improve the robustness of the algorithm，and add batch normalization operation to the TCN residual block to improve the network convergence speed and prevent overfitting. The experimental results show that the accuracy of the proposed TCN perturbation identification model for the identification of various disturbances of vehiclenetwork coupling reaches more than 96.90%，which can more reliably realize the identification of vehicle-network coupling disturbances compared with the training effect of deep convolutional neural network.

      Key words：electrified railways；vehicle-network coupling；disturbances identification；temporal convolutional

neural networks（TCN）；residual block；BN（batch normalization）layer