基于无监督聚类和频繁子图挖掘的电力通信网缺陷诊断与自动派单  

Fault diagnosis and auto dispatchin of power communication network based on unsupervised clustering and frequent subgraph mining

作　　者：吴季桦 朱鹏宇 吴子辰 顾彬 洪涛 郭波 王晶[1] 王敬宇[1] WU Jihua;ZHU Pengyu;WU Zichen;GU Bin;HONG Tao;GUO Bo;WANG Jing;WANG Jingyu(State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China;State Grid Electric Power Research Institute Co.,Ltd,Nanjing 210012,China;Information and Communication Branch of State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210024,China)

机构地区：[1]北京邮电大学网络与交换国家重点实验室,北京100876 [2]国网电力科学研究院有限公司,江苏南京210012 [3]国网江苏省电力有限公司信息通信分公司,江苏南京210024

出　　处：《电信科学》2021年第11期51-63,共13页Telecommunications Science

基　　金：国家电网公司科技项目(No.5700-202040367A-0-0-00)。

摘　　要：缺陷诊断一直是电力通信领域研究的难点之一。基于人工规则的缺陷诊断已经无法应对告警数据的海量增长。基于有监督学习的智能方法需要大量的标注数据和较长的系统构建时间,且大多面向指标性数据,实现部署缺乏可行性。面向告警数据,提出一种基于无监督聚类和频繁子图挖掘实现告警归并和缺陷模式发现的自学习算法,设计了一个自动化完成缺陷诊断及处置的架构。该架构具有良好的可扩展性和迭代更新能力,并部署于实际缺陷自动派单系统中。通过真实场景数据集进行实验验证,结果显示出良好的性能表现,实现了对缺陷的及时发现及精准派单维护。Fault diagnosis is one of the most challenging tasks in power communication.The fault diagnosis based on rules can no longer meet the demand of massive alarms processing.The existing approaches based on the supervised learning need large sets of the labeled data and sufficient time to train models for processing continuous data instead of alarms,which are far behind the feasibility of deployment.As for alarm correlation and fault pattern discovery,a self-learning algorithm based on the density-based clustering and frequent subgraph mining was proposed.A novel approach for automatic fault diagnosis and dispatch were also introduced,which provided the scalable and self-renewing ability and had been deployed to the automatic fault dispatch system.Experiments in the real-world datasets authorized the effectiveness for timely fault discovery and targeted fault dispatch.

关 键 词：电力通信 缺陷诊断 无监督聚类 频繁子图挖掘 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]