Synthetic Oil Chromatographic Sample Generation for Electric Power Transformer Based on Causal GAN

Electric power transformer plays a vital role in ensuring the safe, reliable, and cost-effective electric power system. Researching on fault diagnosis related technology for electric power transformers to accurately identify fault types and characteristics holds immense importance for transformer maintenance and the development of appropriate maintenance strategies. However, in real practice, the lack of oil chromatographic data and the imbalanced percentage of fault case data currently impede the development of the electric power transformer health monitoring. In order to solve the problems, we introduce a generation method using GAN (Generative Adversarial Networks) model combining causal mechanism as conditions. We evaluate our proposed method on the quality of synthetic generated data and the accuracy of fault diagnosis comparing with real data. The results show that the proposed method learns from the data and aligned with real-world data but need further improvement.