基于大数据的时间序列异常点检测研究

发布时间：2018-04-29 00:39

本文选题：异常点检测 + 时间序列　；参考：《计算机技术与发展》2016年05期

【摘要】：针对传统时间序列异常点检测方法在处理大量数据时检测精度与效率低下的缺陷,文中提出一种基于大数据技术的全新时间序列异常点检测方法。首先介绍了传统时间序列异常点检测方法并分析了其缺陷。其次介绍了基于大数据方法的理论推导,包括特征提取、奇异点检测及异常点判别,具体为采用大数据方法将海量序列分解为周期分量、趋势分量、随机误差分量及突发分量四个不同分量,对不同分量进行特征提取并根据特征提取结果进行奇异点检测,并在此基础上利用序列特点判别奇异点是否为异常点。最后通过实验分析对比验证大数据方法的可行性与效率。实验结果表明,基于大数据方法的时间序列异常点检测相比于传统的方法具有更高的检测精度与更快的检测速率。
[Abstract]:Aiming at the defects of the traditional time series anomaly detection method, which is low precision and efficiency in dealing with a large amount of data, a new time series anomaly detection method based on big data technology is proposed in this paper. Firstly, the traditional time series anomaly detection method is introduced and its defects are analyzed. Secondly, the theoretical derivation based on big data's method is introduced, including feature extraction, singular point detection and anomaly discrimination. Four different components of random error component and burst component are used to extract the different components and detect the singularity points according to the feature extraction results. On the basis of this, the singularity points are judged by the sequence characteristics to determine whether the singular points are abnormal points. Finally, the feasibility and efficiency of big data's method are verified by experimental analysis and comparison. The experimental results show that the time series anomaly detection based on big data method has higher detection accuracy and faster detection rate than the traditional method.
【作者单位】：南京邮电大学自动化学院;
【基金】：江苏省自然科学基金(BK20140877,BE2014803)
【分类号】：TP311.13

【相似文献】