Aiming at the problem of data stream anomaly in bridge health monitoring system,this paper proposes a micro cluster-based data stream anomaly detection framework.First,it pre-processes the primitive acquisition data by data merging and missing data imputation. Since there is certain correlation between the data of measuring points of each sensor in monitoring system,it uses principal component analysis to extract bridge’s main feature parameters in order to remove the redundant information.Then it converts the data streams into micro clusters with density clustering algorithm,carries out real-time generation of micro clusters,and maintains the micro clusters according to their updating mechanism,as well as classifies data streams.It is demonstrated through the experiment on monitoring data of a certain bridge in Hubei Province that the proposed method has stronger capability of anomaly detection,and is able to self-adapt for the concept drift phenomenon.%针对桥梁健康监测系统中的数据流异常问题，提出一种基于微簇的数据流异常检测框架。首先对原始采集信号进行数据合并、缺失值填补等预处理；由于监测系统各传感器测点数据间存在一定的关联，利用主成分分析法提取桥梁主要特征参数，去除重叠信息；利用密度聚类算法把数据流转换成微簇，进行微簇的实时生成，并根据微簇更新机制进行微簇维护，对数据流进行分类。通过对湖北某大桥监测数据的实验表明，该方法具有较好的异常识别能力，可以自适应概念漂移现象。
展开▼
