Transformer fault diagnosis method based on MTF and GhostNet

Abstract

To solve the limitations of the DGA technique in transformer fault diagnosis, we propose a transformer fault diagnosis method that combines the MTF conversion, the GhostNetV2, transfer learning, and the optimized SSA algorithm. Firstly, the MTF conversion is applied to convert the 1D DGA data into 2D images that are easier to analyze; then, with the help of the GhostNetV2 that is pre-trained on a large dataset, the transfer learning is implemented to deepen the feature understanding and the GhostNetV2 is fine-tuned to meet the needs of fault classification, and the output layer incorporates the gated recurrent unit network and the multi-head self-attention layer to optimize the diagnostic performance; finally, through the improved sparrow search algorithm that integrates adaptive t-distribution and Levy flight strategy, the parameters are finely optimized to further enhance the accuracy of fault diagnostic. The experimental results show that the proposed method outperforms other methods in evaluation metrics, and significantly improves the accuracy and effectiveness of transformer fault diagnosis.