For improving the fault diagnosis performance of power transformer under the imbalance scenario of dis-solved gas data in oil

a method based on data enhancement and feature augmentation combined with convolutional neural network for transformer fault diagnosis is proposed in this paper. Firstly

a data augmentation method based on conditional tabular generative adversarial network (CTGAN) combined with cascade reject sampling (CRS) is set up to achieve high quality equalization of unbalanced dataset. Then

feature construction and screening method based on an all-type gas ratios with random forests (GRRF) algorithm is constructed to enhance feature dimension and enrich feature diversity. Finally

a fault diagnosis model based on 2D improved convolutional neural network (2D-ICNN) is set up

and the effectiveness of the proposed method is verified through experiments. The results demonstrate that

in comparison to both the oversampling method and CTGAN

the proposed CTGAN-CRS significantly enhances the quality of the generated data. Additionally

the GRRF feature construction method substantially increases feature rich-ness and

on this basis

the fault diagnosis accuracy is improved further by using the improved 2D-ICNN model on this foundation.