基于DTW-kmedoids算法的时间序列数据异常检测  

Anomaly Detection of Time Series Data Based on DTW-Kmedois Clustering Algorithm

作　　者：宗文泽 吴永明[1,2] 徐计 黎旭 王晨[1] ZONG Wen-ze;WU Yong-ming;XU Ji;LI Xu;WANG Chen(Key Laboratory of Modern Manufacturing Technology of Ministry of Education,Guizhou University,Guiyang 550025,China;State Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学现代制造技术教育部重点实验室,贵阳550025 [2]贵州大学公共大数据国家重点实验室,贵阳550025

出　　处：《组合机床与自动化加工技术》2022年第5期120-124,128,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金资助项目(51505094,61962009);贵州省科学技术基金计划项目[(2016)1037];贵州省科技支撑计划项目[(2017)2029]。

摘　　要：针对工业生产中传统聚类算法直接应用于时间序列聚类效果准确性较低的问题,提出一种基于DTW距离度量的K-medoids算法。使用DTW计算时序数据之间的距离取代传统的欧氏距离度量方式,提高了相似性度量算法精度,同时也提高了聚类算法的准确性,并通过构建阈值机制实现了对时间序列数据的监督与异常检测。最后,结合烟叶含水率的时间序列数据进行分析,与传统聚类算法的异常检测模型比较,实验结果表明,DTW-kmedoids算法对时序数据的监督与异常检测具有可靠性、准确性。In order to solve the problem that the accuracy of traditional clustering algorithm applied directly to time series clustering in industrial production is low,a K-medoids algorithm based on DTW distance measurement is proposed.DTW is used to calculate the distance between time series data instead of the traditional Euclidean distance measurement,which improves the accuracy of similarity measurement algorithm and the accuracy of clustering algorithm,and realizes the supervision and anomaly detection of time series data by building a threshold mechanism.Finally,combining with the time series data of tobacco moisture content,and comparing with the anomaly detection model of traditional clustering algorithm,the experimental results show that the DTW-kmedoids algorithm is reliable and accurate in the supervision and anomaly detection of time series data.

关 键 词：时间序列数据 DTW 聚类 异常检测 

分 类 号：TH16[机械工程—机械制造及自动化] TG506[金属学及工艺—金属切削加工及机床]