Abstract: To address the problems of low accuracy for fault diagnosis for track circuits and difficulty in collecting high-quality fault data, a fault diagnosis method based on the combination of one-class support vector machine (OC-SVM) and deep neural network (DNN) is proposed. This method uses OC-SVM model to perform single-class recognition of data. The positive sample data are input into DNN model for training and prediction to label and collect the negative sample data. A large number of experiments are conducted using ZPW-2000R track circuit signal data. Results show that the OC-SVM model can accurately identify the positive and negative sample data, and DNN model can accurately and efficiently diagnose 15 types of data with an accuracy rate of 99%. Compared with the three fault diagnosis methods of particle swarm algorithm optimized support vector machine, convolutional neural network and stacked self-encoder, this combined method has a higher accuracy and more stable diagnosis effectiveness.

Key words: fault diagnosis, deep neural network, one-class support vector machine, ZPW-2000R track circuit