Transformer fault diagnosis based on adversarial generative networks and deep stacked autoencoder

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Transformer fault diagnosis relies on oil chromatogram data, but insufficient and imbalanced samples hinder deep learning model performance.
• Data limitations lead to overfitting, poor model representativeness, and inaccurate fault predictions.

Purpose of the Study:

• To address the challenge of limited and imbalanced transformer oil chromatogram data for deep learning models.
• To enhance the accuracy and reliability of transformer fault diagnosis systems.

Main Methods:

• Utilized a conditional Wasserstein generative adversarial network with gradient penalty optimization (CWGAN-GP) for data expansion.
• Expanded 500 sets of transformer oil chromatography data covering 5 fault types.
• Employed a deep autoencoder for transformer fault classification and compared CWGAN-GP with other generative adversarial network variants.

Main Results:

• The CWGAN-GP method achieved an overall fault diagnosis accuracy of 93.2%, a 4.98% improvement over original imbalanced data.
• The proposed method outperformed other sample expansion techniques, showing accuracy improvements of 1.70%-3.05%.

Conclusions:

• Conditional Wasserstein generative adversarial network with gradient penalty optimization is effective for expanding imbalanced transformer oil chromatogram datasets.
• The enhanced dataset significantly improves deep learning model accuracy for transformer fault diagnosis.