基于异构图注意力的工业设备故障诊断知识图谱补全方法
Knowledge graph completion method for industrial equipment fault diagnosis based on heterogeneous graph attention
马亚杰 1刘洋 1姜斌 1冒泽慧 1刘剑慰 1刘文静2
作者信息
- 1. 南京航空航天大学自动化学院,南京 211106
- 2. 北京控制工程研究所,北京 100190;空间智能控制技术重点实验室,北京 100190
- 折叠
摘要
针对工业设备故障诊断知识图谱故障实体属性残缺、故障关系链接缺失的问题,本文提出了一种基于知识图谱异构图注意力网络(knowledge graph heterogeneous graph attention network,KGHAN)模型的工业设备故障诊断知识图谱补全方法,通过对故障实体概念补全和故障关系链接补全完善了工业设备故障诊断知识图谱.所提KGHAN模型在异构图注意力网络模型的基础上,融合了故障知识结构信息和故障图结构信息,有效地表征了故障实体和故障关系的嵌入表示,提高了故障实体概念补全任务的准确率和故障关系链接补全任务的命中率.将所提工业设备故障诊断知识图谱补全方法应用在国内某企业的工业设备故障运维数据上,结果表明,故障实体概念补全任务的准确率提高了约10％,故障关系链接补全任务的命中率提高了约37％,验证了方法的有效性.
Abstract
To address the issue of a serious lack of fault entity attributes and fault relation links in an industrial equipment fault diagnosis knowledge graph,this paper develops an industrial equipment fault diagnosis knowledge graph completion scheme based on a knowledge graph heterogeneous graph attention network(KGHAN)model.By combining fault knowledge structure information and fault graph structure information in a heterogeneous graph attention network(HAN)model,the developed KGHAN model effectively represents the embedding representations of fault entities and fault relations,which enhances the accuracy of the fault entity concept completion task and the hit rate of the fault relation link completion task.We apply our developed KGHAN model-based industrial equipment fault diagnosis knowledge graph completion scheme to the industrial equipment fault operation and maintenance data of a local enterprise.The results show that the accuracy of the fault entity concept completion task and the hit rate of the fault relation link completion task increased by about 10％and 37％,respectively,which confirms the effectiveness of our method.
关键词
知识图谱补全/知识图谱/图神经网络/故障诊断/工业设备Key words
knowledge graph completion/knowledge graph/graph neural network/fault diagnosis/industrial equipment引用本文复制引用
基金项目
科技创新2023-"新一代人工智能"重大项目(2020AAA0109305)
出版年
2024
NETL
NSTL
万方数据