Voiceprint Recognition Method for Urban Rail Power Transformers Based on Data Augmentation and Multi-Domain Fusion

Urban rail power transformers are key components of the urban rail power supply system, and their fault detection is essential for ensuring the reliability of power delivery. In recent years, voiceprint-based fault detection for transformer have emerged as a promising direction due to their non-contact and non-intrusive nature. Especially, the application of deep learning techniques has led to significant progress in this field. However, challenges such as class imbalance, limited feature diversity, and noise interference still persist. To address these issues, this paper proposes a novel voiceprit recognition method for urban rail power transformers based on data augmentation and multidomain fusion. To validate the effectiveness of the proposed method, experiments were conducted on a dataset provided by a subway power supply department. The results demonstrate that our method achieves 100 % accuracy on the test set, and maintains over 98 % diagnostic accuracy even under high noise conditions of SNR=-4dB, significantly outperforming other baseline methods.