中文摘要：

随着传感器广泛应用于各个领域,在传感器生成的时间序列上识别事件越来越受到广泛的关注。针对震荡的传感器时间序列,提出事件分类算法BEC。对于原始长时间序列和标记时间点作为类标签,BEC主要解决了两个问题。首先是将标记时间点扩展为包含充分信息的子序列以分类,再者是提取基于突变的特征以训练分类模型。实验结果证明,无需大部分时间序列分类问题中不现实的假设和太多人力干预,BEC提取的基于突变的特征能够充分描述事件,极大保留事件中关键信息,在现实数据集上的表现优于现有的时间序列分类算法。

英文摘要：

Detecting events on time series data generated by sensors has received a great amount of attention with increasingly deployment of variable sensors. This paper proposes a novel framework for classifying events upon oscillating data of sensors called BEC. Given a long raw time series and class labels on marked time points, BEC extracts burstbased features to represent events. There are mainly two important tasks to be solved. First, BEC automatically extends labeled time points to appropriate subsequences containing sufficient information. Second, BEC extracts burst-based features to train the classification model. It is demonstrated on real-life datasets that without unrealistic assumptions and human interventions, BEC extracts burst-based features can fully detect the event, greatly retain the key information in the event, and the performance of the actual data set is better than the existing time series classification algorithm.