ConvTasNet Based Transformer Sound Noise Reduction and Condition Recognition Network
- Resource Type
- Conference
- Authors
- Wu, Hao; Lu, Xin; Feng, Yanwei; Zeng, Zuowei; Huang, Zhirong; Hu, Zhaoyu; Li, Zhe
- Source
- 2023 3rd Power System and Green Energy Conference (PSGEC) Power System and Green Energy Conference (PSGEC), 2023 3rd. :965-969 Aug, 2023
- Subject
- Aerospace
Computing and Processing
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Transportation
Employee welfare
Substations
Working environment noise
Noise reduction
Neural networks
Green products
Transformers
acoustic recognition
acoustic noise reduction
sound source separation
convolutional neural network
- Language
This paper proposes a transformer audio noise reduction and condition recognition network based on ConvTasNet to improve the recognition accuracy of transformer sound patterns in the presence of environmental noise. The network employs ConvTasNet for audio noise reduction and a convolutional neural network for transformer condition recognition. A dataset of audio data under different operating conditions of the transformer is prepared through simulation experiments. The noise reduction and condition recognition network is compared with other noise reduction methods in terms of noise reduction effect and recognition accuracy improvement. Results demonstrate that the noise reduction and condition recognition network outperforms other noise reduction methods, achieving a 9.84 dB improvement in SISNR of the audio and 25.85% improvement in recognition accuracy. In field applications, the proposed network effectively improves the recognition accuracy of transformer acoustic patterns.