基于图知识推理的输电线路缺销螺栓识别方法

PDF下载HTML

Title:

Pin-missing bolts recognition method for transmission lines based on graph knowledge reasoning

作者:

赵振兵^1,2,3, 王睿¹, 赵文清^2,4, 张珂^1,3, 翟永杰⁴

1. 华北电力大学 电子与通信工程系，河北 保定 071003;
2. 华北电力大学 复杂能源系统智能计算教育部工程研究中心，河北 保定 071003;
3. 华北电力大学 河北省电力物联网技术重点实验室，河北 保定 071003;
4. 华北电力大学 控制与计算机工程学院，河北 保定 071003

Author(s):

ZHAO Zhenbing^1,2,3, WANG Rui¹, ZHAO Wenqing^2,4, ZHANG Ke^1,3, ZHAI Yongjie⁴

1. Department of Electronic and Communication Engineering, North China Electric Power University, Baoding 071003, China;
2. Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, North China Electric Power University, Baoding 071003, China;
3. Hebei Key Laboratory of Power Internet of Things Technology, North China Electric Power University, Baoding 071003, China;
4. School of Control and Computer Engineering, North China Electric Power University, Baoding 071003, China

关键词:

输电线路; 螺栓; 缺销识别; 图知识推理; 知识表达; 标签依赖; 图卷积网络; 类别表示

Keywords:

transmission line; bolt; pin-missing recognition; graph knowledge reasoning; knowledge expression; tag dependencies; graph convolutional network; category representation

分类号:

TP18

DOI:

10.11992/tis.202205004

摘要:

针对输电线路缺销螺栓识别任务中存在的视觉不可分与样本不平衡问题，提出了基于图知识推理的输电线路缺销螺栓识别方法。首先通过知识表达模块学习到各类螺栓有判别力特征的类别表示；然后深入挖掘螺栓数据集中螺栓种类之间的相关性，提取出标签共现信息；最后将类别表示作为输入特征，以静态图和动态图的相关概率矩阵表征标签共现信息，通过知识推理模块完成图知识的传播与增强，从而实现缺销螺栓的识别。在实验阶段，将所选取的3类金具上的6种螺栓作为实验对象。对比实验结果表明，本文方法对缺销螺栓的识别效果优于其他方法，较原始模型提升了9.13%的准确率。消融实验结果表明，本文所提取的螺栓类别表示、静态图及动态图信息均能够提升缺销螺栓的识别效果。

Abstract:

Aiming to address the challenges of visual indistinguishability and sample imbalance in identifying missing bolts in transmission lines, we propose a method that relies on graph knowledge reasoning. First, the knowledge expression module learns class representations of various bolts with discriminative characteristics. Then, the label co-occurrence information is extracted through the analysis of correlations among bolt types in the bolt dataset. Finally, the knowledge inference module performs dissemination and enhancement of graph knowledge by taking the category representation as an input feature and using static and dynamic graphs, representing label co-occurrence information, as adjacency matrices. This leads to the identification of missing bolts. In the experimental phase, six bolt types on three selected fittings were selected as test subjects. The results show that our method outperforms other methods in identifying missing bolts, improving accuracy by 9.13% compared to the original model. The results of the ablation experiment demonstrate that the category representation, static graph, and dynamic graph information of bolts extracted in our method all contribute to improved recognition of missing bolts.