Research On Infrared Fault Detection Of Substation Equipment Based On Deep Learning
IOSR Journal of Electronics and Communication Engineering · 2026-02 · 21 (2) · 05-13
DOI: 10.9790/2834-2102010513 ↗Kurz gefasst
Fachartikel in der Zeitschrift IOSR Journal of Electronics and Communication Engineering, 2026-02. Verfasst von Ruxin Gao und 1 weiteren Autor:innen.
Abstract (Originalsprache)
The reliable operation of power systems cannot be achieved without efficient and precise operation and maintenance methods. Based on the problems existing in actual operation and maintenance, and starting from the demand orientation, this paper focuses on the characteristics of low efficiency, reliance on judgment experience, and uncertainty of manual inspection in substations, and proposes a target detection method based on deep learning and YOLOv7. This method is an improvement on the traditional target detection. Through ablation and comparison experiments, the improved model can enhance detection accuracy, reduce the network model, and increase detection speed compared with the traditional model.
Background
As an important networking component in the operation of the power grid, substations can achieve the conversion between different voltage levels and the distribution of loads. Undoubtedly, they play a crucial role in the safety of the power grid and the reliability of power supply. However, with the aging of substation equipment and the reduction of its service life, the failure rate of the equipment gradually increases, intensifying the difficulty of substation maintenance work. In addition, as most of the main grid equipment in substations is set outdoors and located in complex urban environments or remote rural areas, the difficulty of fault handling is even greater, which brings great pressure and safety hazards to power enterprises. Conducting regular inspections of substation equipment as well as periodic and special safeguard inspections is an important means to ensure that power grid enterprises have the ability to issue risk warnings and eliminate risks in advance. Therefore, how to detect the faults and defects of substation equipment as early as possible to prevent substation accidents in advance has become a focus of attention in the power industry. Materials and methods: By conducting infrared photography of the operating equipment in the substation, and then preliminarily processing the data for data augmentation and classification, an infrared detection dataset for the substation was created. The object detection method of YOLOv7 is adopted. Firstly, the GhostNetV2 module is introduced to conduct a lightweight design for the feature extraction backbone of YOLOv7. Aiming at the problem of insufficient fusion of the feature layer of the YOLOv7 algorithm model, the bidirectional Feature Pyramid Network (BiFPN) feature fusion module is introduced to improve the detection accuracy. Then, the loss function of YOLOv7 is replaced with the SIOU loss function to achieve dynamic adjustment of the regression loss of the model prediction box, thereby accelerating the convergence speed. And it plays a role in improving the recognition accuracy of the model. Finally, the model was experimented on the infrared detection dataset of the substation and compared with the mainstream methods to verify the effectiveness of the method in this chapter.
Result
The experimental results show that the substation infrared image defect detection model designed in this chapter has a 2.3% improvement in detection accuracy compared to YOLOv7, while the model size is reduced by 32%. When the model size is basically the same as that of YOLOv8m, the detection accuracy is increased by 4% and the detection speed by 43%. It can be concluded from this that the improved YOLOv7 model has achieved relatively ideal improvements in both defect detection accuracy and detection speed.
Zitieren
APA
Gao, R., & Ding, H. (2026). Research On Infrared Fault Detection Of Substation Equipment Based On Deep Learning. IOSR Journal of Electronics and Communication Engineering, 21(2), 05-13. https://doi.org/10.9790/2834-2102010513
Vancouver
Gao R, Ding H. Research On Infrared Fault Detection Of Substation Equipment Based On Deep Learning. IOSR Journal of Electronics and Communication Engineering. 2026;21(2):05-13. doi:10.9790/2834-2102010513
BibTeX
@article{Gao2026,
author = {Gao, Ruxin and Ding, Hao},
title = {Research On Infrared Fault Detection Of Substation Equipment Based On Deep Learning},
journal = {IOSR Journal of Electronics and Communication Engineering},
year = {2026},
volume = {21},
number = {2},
pages = {05--13},
doi = {10.9790/2834-2102010513}
}
EUVIMED