Abstract: Time series event detection is an important task in the field of time series data mining which starts from time series data,and the event information in the data is detected by data mining method.Event is the basis of process mining,which can be used to discover,analyze and optimize processes.Two abstract methods were used to complete the event detection.The time series was divided into state intervals according to the given label,then the frequent time patterns were found in the state interval sequence,and the possible event points were obtained by screening the frequent patterns through association rules.The feasibility and efficiency of the algorithm were verified by experiments on simulated data sets and real data sets.

Key words: time series, event detection, multi-resolution analysis, wavelet analysis, data mining

摘要： 鉴于时间序列事件检测为时间序列数据挖掘领域的重要任务之一,以时间序列数据为输入,通过数据挖掘方法检测出数据中存在的事件信息,将其作为流程发现、合规检测和优化等下游任务的基础。采用两次抽象的方法完成事件检测,即根据给定标签将时间序列划分为状态区间,在状态区间序列中发现频繁出现的时间模式,通过关联规则对频繁模式进行筛选得到可能的事件点。通过模拟数据集和真实数据集上的实验验证了算法的可行性和运行效率。

关键词: 时间序列, 事件检测, 多尺度分析, 小波分析, 数据挖掘