State Identification of Transformer Under DC Bias Based on Wavelet Singular Entropy

Jun Liu, Huarong Zeng, Wei Niu, Peilong Chen, Kui Xu, Peng Zeng, Lu Zhao, Sheng LinIEEE Access 2021 · DOI

Abstract

To identify the DC bias state of transformers, a DC bias state identification method of transformer based on wavelet singular entropy is proposed in this paper. The vibration principle of transformers under DC bias has been analyzed. By combining continuous wavelet transform, singular value decomposition, and information entropy, the analysis method of wavelet singular entropy is proposed. The vibration signal of the transformer before and after DC bias is transformed by continuous wavelet transform, and then the wavelet time-frequency diagram is compared and analyzed. Furthermore, the DC bias state of the transformer is identified by wavelet singular entropy of vibration signal. The wavelet singular entropy under different states is in different numerical ranges, and the wavelet singular entropy of vibration signal under DC bias is significantly greater than that under no DC bias. The wavelet singular entropy effectively reflects the DC bias state of transformers. Then the proposed method is applied to a 500 kV transformer and a 220 kV transformer in the China Southern Power Grid. The results show that the proposed method can accurately and effectively identify the DC bias state of transformers.

Introduction: The Challenge of DC Bias

Power transformers are critical components of the electrical grid, yet they are increasingly vulnerable to DC bias. This phenomenon occurs when direct current flows into a transformer's neutral point, often triggered by geomagnetic storms, high-voltage DC transmission systems, or stray currents from urban rail transit.

When DC bias occurs, it shifts the magnetic flux in the iron core, leading to half-cycle saturation. This saturation results in local overheating, increased noise, and intense mechanical vibrations. Identifying this state is traditionally difficult because measuring excitation currents directly is technically challenging for transformers in active service.

Magnetostriction characteristics under normal excitation vs. DC bias.

Methodology: Wavelet Singular Entropy (WSE)

The proposed method utilizes non-invasive vibration sensors mounted on the transformer's outer tank. The core idea is that mechanical vibration serves as a reliable proxy for internal electromagnetic states. The analysis follows a three-step mathematical pipeline:

  1. Continuous Wavelet Transform (CWT): The raw vibration signal is decomposed into time and frequency components using the db4 wavelet. This handles the non-stationary nature of transformer vibrations.
  2. Singular Value Decomposition (SVD): SVD is applied to the wavelet matrix to extract essential features while filtering out background noise.
  3. Information Entropy: Shannon’s entropy formula is applied to the singular values to calculate Wavelet Singular Entropy (WSE).
Typical vibration signal from a transformer tank used for analysis.

Under normal conditions, vibration energy is concentrated in specific low-frequency bands, resulting in low entropy. However, DC bias causes core saturation, creating a complex vibration pattern across many frequency multipliers, which manifests as a significant increase in WSE.

Field Validation and Results

The method was validated using field data from the China Southern Power Grid, specifically targeting 500 kV and 220 kV main transformers.

  • Numerical Separation: In 500 kV transformer tests, normal state WSE fluctuated between 1.6 and 1.7. During verified DC bias (neutral current ~11A), the WSE rose sharply to 2.1–2.2.
  • Identification Threshold: Based on these results, a threshold of $M = 1.8$ was established. Values exceeding this indicate a DC bias state.
  • Frequency Distribution: Qualitative analysis confirmed that while normal vibrations are mostly below 0.4 kHz, DC bias significantly increases activity in the 0.4 kHz to 1 kHz range.

Conclusion

The Wavelet Singular Entropy method provides a robust, non-invasive metric for monitoring transformer health. By measuring the "disorder" of tank vibrations, it accurately distinguishes between normal operation and hazardous DC bias. Future work will focus on automating threshold calibration for different transformer classes and assessing the impact of extreme environmental noise.

Summary updated

Abstract

To identify the DC bias state of transformers, a DC bias state identification method of transformer based on wavelet singular entropy is proposed in this paper. The vibration principle of transformers under DC bias has been analyzed. By combining continuous wavelet transform, singular value decomposition, and information entropy, the analysis method of wavelet singular entropy is proposed. The vibration signal of the transformer before and after DC bias is transformed by continuous wavelet transform, and then the wavelet time-frequency diagram is compared and analyzed. Furthermore, the DC bias state of the transformer is identified by wavelet singular entropy of vibration signal. The wavelet singular entropy under different states is in different numerical ranges, and the wavelet singular entropy of vibration signal under DC bias is significantly greater than that under no DC bias. The wavelet singular entropy effectively reflects the DC bias state of transformers. Then the proposed method is applied to a 500 kV transformer and a 220 kV transformer in the China Southern Power Grid. The results show that the proposed method can accurately and effectively identify the DC bias state of transformers.
BibTeX
@inproceedings{liu2021state,
  author = {Jun Liu and Huarong Zeng and Wei Niu and Peilong Chen and Kui Xu and Peng Zeng and Lu Zhao and Sheng Lin},
  title = {{State Identification of Transformer Under DC Bias Based on Wavelet Singular Entropy}},
  booktitle = {IEEE Access},
  year = {2021},
  doi = {10.1109/access.2021.3072913},
}
Figure