ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2018, Vol. 55 ›› Issue (3): 524-536.doi: 10.7544/issn1000-1239.2018.20170804

所属专题: 2018边缘计算专题

• 网络技术 • 上一篇    下一篇



  1. (山东大学计算机科学与技术学院 济南 250101) (
  • 出版日期: 2018-03-01
  • 基金资助: 

Edge Computing Application: Real-Time Anomaly Detection Algorithm for Sensing Data

Zhang Qi, Hu Yupeng, Ji Cun, Zhan Peng, Li Xueqing   

  1. (School of Computer Science & Technology, Shandong University, Jinan 250101)
  • Online: 2018-03-01

摘要: 随着物联网技术的不断发展,已逐步进入“万物互联”的新时代.针对物联网中实时采集的传感数据总体质量低下的问题,提出基于边缘计算的传感数据异常实时检测算法.该算法首先对相应的传感数据以“时间序列”的形式进行表示,并建立基于边缘计算的分布式传感数据异常检测模型;其次利用单源时间序列自身的连续性以及多源时间序列之间的相关性,分别对实时传感数据中出现的数据异常进行有效检测,并分别形成相应的异常检测结果集;最后将上述2个异常检测结果集进行有效地融合处理,从而得到更加准确的异常数据检测结果.通过实验验证该算法的检测准确性和有效性,结果显示:该算法检测时间短并且异常检出率高.

关键词: 物联网, 边缘计算, 时序数据, 相关性, 离群距离

Abstract: With the rapid development of Internet of things (IoT), we have gradually entered into the IoE (Internet of everything) era. In face of the low quality of real-time gathering sensor data in IoT, this paper proposes a novel real-time anomaly detection algorithm based on edge computing for streaming sensor data. This algorithm firstly expresses the corresponding sensor data in the form of time series and establishes the distributed sensing data anomaly detection model based on edge computation. Secondly, this algorithm utilizes the continuity of single-source time series and the correlation between multi-source time series to detect anomaly data from streaming sensor data effectively and respectively. The corresponding anomaly detection result sets are also generated in the same process. Finally, the above two anomaly detection result sets would be effectively fused in a certain way so as to obtain more accurate detection result. In other words, this algorithm achieves a higher detection rate compared with other traditional methods. Extensive experiments on the real-world dataset of household heating data from the Jinan municipal steam heating system, which collects monitoring data from 3084 apartments of 394 buildings, have been conducted to demonstrate the advantages of our algorithm.

Key words: Internet of things, edge computing, time series data, correlation, outlier distance