Research on Transformer Fault Diagnosis Using Boundary Equilibrium Generative Adversarial Network for Data Augmentation

The quantity and quality of fault samples in power transformers significantly impact the diagnosis performance of dissolved gas analysis (DGA) based fault diagnosis approaches. This paper addresses the issues of insufficient data sample and data imbalance in power transformer fault diagnosis by proposing a data augmentation method based on Boundary Equilibrium Generative Adversarial Network (BEGAN). Firstly, we analyze the characteristics of transformer fault samples and the impact of data imbalance on diagnostic models. By comparing with common data augmentation methods, we demonstrate the advantages of BEGAN in generating fault samples and introduce multiple metrics to evaluate the quality of the generated data. Then a BEGAN-based fault data generation model is constructed, with detailed discussion on parameter settings and structural design. Lastly, by incorporating the generated data of varying quantities into the training set of the diagnostic model, we validate that BEGAN can effectively generate high-quality transformer fault samples, thus improving the quality of the training set and enhancing the accuracy and robustness of the fault diagnosis model. Our research provides a novel approach for the data scarcity or imbalance problem in power transformer fault diagnosis, which is of significant importance for improving the diagnosis accuracy.