Affiliations: [a]
Department of Production Technology, Qinzhou Power Supply Bureau of Guangxi Power Grid Co., Ltd., Qinzhou, China
| [b] Qinzhou Power Supply Bureau Urban Branch, Qinzhou Power Supply Bureau of Guangxi Power Grid Co., Ltd., Qinzhou, China
Correspondence:
[*]
Corresponding author. Deyin Wang, Department of Production Technology, Qinzhou Power Supply Bureau of Guangxi Power Grid Co., Ltd., Qinzhou, 535000, China. Email: wdy_15994607960@163.com.
Abstract: Foreign objects identification in the distribution network is an important link in the security of electric power, and is of great significance to the normal transportation of electric power. At present, a lot of equipment in the distribution network is in the open air environment, facing a large number of foreign interference. These foreign objects not only bring potential safety hazards to the distribution network, but also easily lead to short circuit, causing power supply difficulties within the region. Therefore, the research first constructs an optimized triplet feature learning model. On this basis, the HOG-SVM depth feature recognition model is proposed. In HOG-SVM, AM is introduced to improve recognition accuracy. In addition, the research enhances the night vision ability of the model by standardizing the features in the image region block. The results show that the AP of the model is stable at more than 90.54%, the average FPR is 2.21%, and the average FNR is 3.17%. The performance of HOG-SVM is significantly better than that of traditional SVM. It verifies the contribution of this research in the field of foreign object recognition and application value in ensuring the security of distribution network.
Keywords: Distribution network, foreign objects, depth characteristics, attention
