传感器网络时间序列数据的事件分类研究

目前智能环境中传感器网络所采集的海量数据面临着进行有效事件的模式分类及异常检测的难题.为了有效对智能环境中传感器网络采集的时间序列数据所表征的事件进行分类,提出了基于协方差特征空间映射数据的聚类分析方法.通过对采集得到的时间序列数据按时隙进行划分,映射到协方差特征空间,然后对映射后的数据进行了动态密度聚类,从而实现对事件的分类;并根据聚类结果建立分类模板,作为对日常事件进行分类划分的检测方法,同时利用所得的分类模板,实现对异常事件的检测.实验结果表明,基于协方差特征空间映射数据的聚类分析方法能有效对传感器网络采集的时间序列数据所表征的事件进行分类,并能有效提升异常事件的检测及筛选效果.

Research of time series data events classification in sensor networks

The big data collected by sensor networks in intelligent environments are faced with the problem of pattern classification and anomaly detection of time series data events. For the efficient event classification of time series data collected by sensor networks, the clustering analysis in covariance feature space was proposed. By partitioning the time series data which were collected by sensor networks, then mapping to the covariance feature space and using density-based clustering algorithms, the classification template was built to classify the usual event. The classification template was used in anomaly detection. The experiments results show that the proposed methods can effectively classify the time series data events, and enhance the performance of anomaly detection.