An Automatic Detection Framework for Electrical Anomalies in Electrified Rail Transit System

Automatic electrical anomaly detection for high-speed rail (HSR) is essential to ensure the safe and efficient operation of HSR. In this article, a novel anomaly detection method is proposed to rapidly identify typical anomalies and detect unknown anomalies in train-grid electrical coupling system (TGECS). The framework of the method consists of feature extraction, pre-classification and re-identification. First, the raw electrical signal is processed by Hilbert transform (HT) to obtain the instantaneous envelope. Second, the one-dimensional (1-D) convolutional neural network (CNN) is adopted to learn deeper features and perform pre-classification. To reduce the number of parameters, the global average pooling layer is used instead of the fully connected layers. To further improve the recognition accuracy, the additive angular margin named arcface is used to enhance the discriminative power of the model. Finally, the re-identification operation determines the final result by calculating the distances between the features. The accuracy of the proposed method for typical electrical anomaly recognition is beyond 99% and for unknown anomaly detection exceeds 86%. The experimental results demonstrate the effectiveness and the superior performance of the proposed method in terms of recognition accuracy, unknown anomalies detection and computational complexity, compared to the NN, 2-D CNN, LSTM and CNN-LSTM.