Research on Turnout Fault Identification Method Based on Improved Convolutional Neural Network

The identification of turnout faults is of great significance in railway unmanned driving systems. It not only ensures the safety of train operations, improves operational efficiency, and reduces maintenance costs, but also promotes the intelligent development of railway systems. Therefore, this paper constructs a turnout fault identification model based on an improved Convolutional neural network, adds a batch normalization processing layer based on a traditional Convolutional neural network, and optimizes the Convolutional neural network structure and network parameters to improve the identification accuracy of Convolutional neural network; In view of the problem that the number of turnout fault data samples is insufficient, according to the characteristics of turnout action power curve sampling data, the tabular data Generative adversarial network (CTGAN) model is constructed, and the comparative analysis between Variational autoencoder VAE and conditional table Generative adversarial network CTGAN is carried out. The results show that the CTGAN application in turnout fault data expansion is reasonable. The simulation comparison experiments of RF and MLP classification algorithms and improved Convolutional neural network, as well as the comparison experiments of Convolutional neural network before and after the improvement, show that the method proposed in this paper has good fault identification accuracy.