A Sample Synthesis Method for Railway Track Circuit Fault Diagnosis Based on Variational Autoencoder and Improved SMOTEENN

Accurately identifying fault modes in track circuits is crucial for ensuring the safe and reliable operation of the equipment. However, in practical diagnostic processes, there is often an imbalance problem where the number of faulty samples is significantly smaller than the number of normal samples. To address this issue, this paper proposes a sample synthesis technique that combines variational autoencoder (VAE) with an improved synthetic minority over-sampling technique edited nearest neighbor (SMOTEENN). At first, VAE is employed to thoroughly extract features from the raw monitoring data. The VAE-encoded data is, then, subjected to hybrid sampling using the SMOTEENN algorithm. Next, a filtering method based on the Chebyshev distance is proposed to further improve the quality of the synthesized samples. The effectiveness of the proposed method is validated using a track circuit simulation dataset. The experimental results demonstrate that the synthesis method proposed in this paper significantly enhances the accuracy of fault diagnosis compared to the SMOTE, SMOTEENN and VAE-SMOTEENN methods.