基于W-ReLU的设备多工况状态异常检测方法  

Equipment Abnormal Detection Method for Dynamic Working Conditions Based on W-ReLU

作　　者：张翼英[1] 王鹏凯 柳依阳 武延年 郭晓艳 ZHANG Yiying;WANG Pengkai;LIU Yiyang;WU Yannian;GUO Xiaoyan(College of Artificial Intelligence,Tianjin University of Science&Technology,Tianjin 300457,China;China Gridcom Co.,Ltd.,Shenzhen 518109,China;Information and Communication Company,State Grid Tianjin Electric Power Company,Tianjin 300000,China)

机构地区：[1]天津科技大学人工智能学院,天津300457 [2]深圳市国电科技通信有限公司,深圳518109 [3]国网天津市电力公司信息通信公司,天津300000

出　　处：《天津科技大学学报》2022年第5期63-70,共8页Journal of Tianjin University of Science & Technology

基　　金：国家自然科学基金青年基金项目(61807024)。

摘　　要：针对光通信设备在多工况条件下由于健康状态的不确定性导致难以及时发现运行异常的问题,对基于监控日志数据量化分析的设备异常检测方法进行研究,提出了基于加权线性修正函数(W-ReLU)的设备多工况状态异常检测方法.首先采用滑动时间窗口机制对告警日志进行划分,并根据每条日志的类型,将日志子集表征为日志特征向量;然后对健康状态下的样本点进行密度峰值聚类分析,以构建设备的基准健康状态矩阵;最后采用W-ReLU非线性映射模型量化评估设备的异常度,并据此进行异常检测.结果表明,与现有的相似性度量方法相比,该方法具有更高的准确率和稳定性.An equipment abnormal detection method based on quantitative analysis of monitoring log data was studied to address the problem that it is difficult to find the abnormal status of optical communication equipment in time due to the uncertainty of the equipment health status under dynamic working conditions.To this end,in this article we propose an equipment abnormal detection method for dynamic working conditions based on weighted rectified linear function(W-ReLU).In this method,the alarm logs were first divided by sliding time window mechanism,and log subsets were represented as log feature vectors based on the type of each log.Then,cluster analysis of density peaks was performed on the sample points under normal operation conditions to construct the baseline health state matrix.Finally,W-ReLU was used to quantitatively evaluate the anomaly degree of the equipment and perform anomaly detection accordingly.Experimental results show that the proposed method has higher accuracy and stability than the traditional similarity measurement methods.

关 键 词：光通信设备 异常检测 特征权重 样本相似性 

分 类 号：TN929[电子电信—通信与信息系统] TP181[电子电信—信息与通信工程]