基于时空相关性的多传感器数据异常检测  

OUTLIERS DETECTION OF MULTI-SENSOR DATA BASED ON SPATIAL-TEMPORAL CORRELATION

作　　者：柳月强 张建锋[1] 祝麒翔 杨会君[1] Liu Yueqiang;Zhang Jianfeng;Zhu Qixiang;Yang Huijun(College of Information Engineering,Northwest A&F University,Yangling 712100,Shaanxi,China)

机构地区：[1]西北农林科技大学信息工程学院,陕西杨凌712100

出　　处：《计算机应用与软件》2020年第10期85-90,共6页Computer Applications and Software

基　　金：陕西省重点研发计划项目(2019NY-164)。

摘　　要：针对环境检测系统中多传感器数据同时具有时间和空间的相关性但在进行异常检测中往往只从单一角度解决的问题,将最近邻检测与聚类检测的思想相结合,提出一种基于时空相关性的异常检测算法(ODSTC)。采用动态阈值的K近邻距离差算法(KNND)实现数据集在时间序列上的异常检测;采用参数优化后的DBSCAN聚类算法(PO_DBSCAN)实现多密度的空间聚类异常检测。采用校试验示范站信息服务体系项目数据集进行实验,并与DBSCAN和PO_DBSCAN等算法进行对比实验。结果表明,该算法的检测率可达98.5%,且时间性能也有所提升。In the environmental detection system,multi-sensor data has temporal and spatial correlation at the same time,but in the process of outliers detection,it is often solved from a single point of view.To solve the problem,we propose an outlier detection method based on spatial-temporal correlation(ODSTC),which combines nearest neighbor detection with cluster detection.The k-nearest neighbor distance algorithm(KNND)based on dynamic threshold was used to detect the outliers in time series.Then the parameter optimized DBSCAN clustering algorithm(PO_DBSCAN)was used to realize the multi-density spatial clustering outlier detection.The experiment used the data set of the information service system project of the University demonstration station,and compared it with DBSCAN and PO_DBSCAN.The results show that the detection rate of our method can reach 98.5%,and the time performance of the algorithm is also improved.

关 键 词：数据分析 异常检测 K近邻距离差 DBSCAN聚类 时空相关性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]