摘要
针对水文行业对数据异常模式检测的实时性要求,提出一种基于特征向量的两阶段异常检测方法。先提取时间序列特征形成符号化的特征向量,再使用改进的K-means方法进行聚类,最后用改进的INN对聚类结果进行评估并将聚类后得到的类簇设成相应特征模型。实验表明,该方法实现了对字符串序列的高效准确的聚类,有效检测出异常模式。
Considering the real-time requirement of data anomaly detection in hydrological industry,a two-stage anomaly detection method based on eigenvector is proposed.Firstly,ESAX method is used to extract time series features to form symbolized feature vectors,then improved K-means method is used to cluster feature vectors.Finally,improved INN is used to evaluate the clustering results and set the cluster after clustering into the corresponding feature model.Experiments show that this method not only effectively realizes efficient and accurate clustering of string sequences,but also effectively detects abnormal patterns in hydrological time series.
作者
张日新
朱跃龙
万定生
毛燠锋
ZHANG Ri-xin;ZHU Yue-long;WAN Ding-sheng;MAO Yu-feng(College of Computer and Information,Hohai University,Nanjing 211100,China;NR Electric Co.,Ltd.,Nanjing 211100,China)
出处
《信息技术》
2019年第11期67-71,77,共6页
Information Technology
基金
国家重点研发计划(2018YFC1508100,2018YFC0407900)
关键词
异常检测
特征聚类
水文时间序列
符号化
anomaly detection
feature clustering
hydrological time series
symbolization