Abstract: It is an important prerequisite for analyzing and avoiding electrical accidents to accurately identify the electrical coupling anomalies of vehicle network in electrified railways. At present, the research on electric coupling anomaly of vehicle network focuses on harmonic resonance, low frequency oscillation, inrush current and short circuit. In the existing identification methods, each algorithm only identifies one type of exception, which cannot cope with the frequent occurrence of multiple exception types in practice. This paper proposed a method for identifying abnormal electrical phenomena (AEP) based on convolutional neural network (CNN), which can monitor the above four typical AEP while distinguishing normal data. Firstly, the characteristics of AEP were analyzed. The waveform structure similarity between different AEP was calculated. According to the characteristics and distribution of the waveform of abnormal phenomena, the CNN suitable for the identification of AEP was proposed. Two parallel feature extraction submodels were used to capture waveform features of voltage and current at the same time, and the extracted features were fused in the Flatten layer. In addition, the original signal was subsampled and the Batch normalization layer was added into the network model to accelerate the convergence speed of the network and avoid overfitting. The experimental results show that precision accuracy of the proposed algorithm for abnormal phenomena reaches 98.17%. And the comprehensive recognition accuracy for the locomotive electrical coupling state reaches 95.75%. The algorithm realizes the automatic identification of AEP in electrified railway.