摘要
时间序列数据是按照时间顺序在不同的时间点采集的数据,反映了某一对象随时间的变化状态和程度。由于时间序列的海量性及复杂性,我们采用频域表示时间序列,并以此为基础提出了基于相似性分析的时间序列异常检测方法。将动态模式匹配距离作为衡量相似性的指标,计算每一个模式同其余各模式之间的相似性,据此确定异常状态。该方法大大降低了数据搜索复杂度,提高了系统效率与准确度。
Time series data are collected at different time points according to the time order to reflect an objective variation states and contents with time.This paper took frequency domain to express the time series in consideration of the magnanimity and complexity of time series and propose the detection method on abnormal time series on account of similarity analysis to take a dynamic pattern matching distance as a index to calculate the similarity between every model and others hereby to ensure the abnormal state.This method could greatly reduce the complexity in data search and improve the efficiency and accuracy of the system.
作者
孙焱
林意
SUN Yan;LIN Yi(School of Digital Media/Jiangnan University, Wuxi 214000, China)
出处
《山东农业大学学报（自然科学版）》
CSCD
2017年第2期287-292,共6页
Journal of Shandong Agricultural University：Natural Science Edition
关键词
时间序列
相似性分析
动态模式匹配
异常检测
Time series
similarity analysis
dynamic pattern matching
abnormal detection
