Transformer fault identification method based on GASF‐AlexNet‐MSA transfer learning

Abstract

The transformer is an important part of the power system and ensures the stable operation of the power grid and electricity safety key equipment. With the increase in electricity demand, it is of great significance to ensure the safe and reliable operation of transformers. However, the commonly used dissolved gas analysis (DGA) method in oil for transformer fault identification has significant drawbacks, so this paper proposes a transformer fault identification method based on GASF‐AlexNet‐MSA transfer learning. The use of GASF to convert one‐dimensional dissolved gas analysis (DGA) data into two‐dimensional images, thus enhancing the comprehensiveness of data representation; the utilization of a pre‐trained AlexNet model through transfer learning, which enables the method to efficiently extract complex features such as textures, shapes, and edges; and the introduction of multiple self‐attention mechanisms that further refine the feature extraction and focuses on the key features, thereby improving the accuracy of fault identification. The proposed model achieves a remarkable accuracy of 97.04% on the publicly DGA dataset, which is 5.19% higher than AlexNet, 6.48% higher than VGG16, 6.12% higher than GoogLeNet, 2.41% higher than ResNet, and 3.71% higher than MobileNet. These results underscore the model's strong feature extraction capabilities and its superior performance in transformer fault identification, providing a valuable reference for enhancing the reliability and safety of power systems.