A Transformer Fault Diagnosis Method Integrating GAF-MViT V3 and OGJO

By combining Gram corner field transformation and MobileViT V3 network structure with OGJO optimization algorithm, the problem of low accuracy in transformer fault diagnosis is solved, and efficient and accurate fault identification and visualization analysis are achieved.

CN122368699APending Publication Date: 2026-07-10SHENYANG LIGONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHENYANG LIGONG UNIV
Filing Date
2026-04-07
Publication Date
2026-07-10

AI Technical Summary

Technical Problem

Existing transformer fault diagnosis methods have low accuracy when dealing with complex operating conditions and mixed faults. Furthermore, traditional methods rely on expert experience, resulting in strong subjectivity and insufficient generalization ability.

Method used

Gram-angle field (GAF) is used to convert one-dimensional fault data into two-dimensional images. Combined with the MobileViT V3 network structure and OGJO optimization algorithm, global and high-frequency features are extracted through a dual-branch network, and the t-SNE algorithm is used for visualization diagnosis.

Benefits of technology

It significantly improves the accuracy of transformer fault diagnosis, with Precision, Recall and F1 values ​​reaching 90.46%, 91.22% and 96.91% respectively, which are better than traditional single-branch models and advanced network models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122368699A_ABST
    Figure CN122368699A_ABST
Patent Text Reader

Abstract

This invention provides a transformer fault diagnosis method integrating GAF-MViT V3 and OGJO, belonging to the field of transformer fault diagnosis technology. First, Gram angle field (GAF) is used to transform the one-dimensional DGA data of the transformer into two-dimensional GASF and GADF images, respectively representing the global trend and local abrupt change characteristics of the signal. Second, a dual-branch network based on MobileViT V3 is constructed. The GASF branch uses the CNN branch of MobileViT V3 to extract global low-frequency features, while the GADF branch uses the Transformer branch to capture local high-frequency features. Dynamic fusion of the two types of features is achieved through a feature recombination module. Finally, the improved OGJO algorithm with a reverse learning strategy is used to globally optimize the hyperparameters, improving the model's generalization ability. The t-SNE algorithm is used to visualize the fused features, enabling interpretable analysis of the diagnostic results.
Need to check novelty before this filing date? Find Prior Art