物联网学报 ›› 2018, Vol. 2 ›› Issue (4): 14-21.doi: 10.11959/j.issn.2096-3750.2018.00076

所属专题: 边缘计算

• 理论与技术 • 上一篇    下一篇

基于边缘计算的物联网监测系统中利用自编码神经网络实现的异常检测

于天琪1,朱咏絮2,王现斌1   

  1. 1 西安大略大学电气与计算机工程学院,安大略 伦敦N6A 5B9
    2 拉夫堡大学沃尔夫森机械、电气与制造工程学院,拉夫伯勒LE11 3TU
  • 修回日期:2018-10-29 出版日期:2018-12-01 发布日期:2019-01-03
  • 作者简介:于天琪(1991-),女,西安大略大学电气与计算机工程学院博士生,主要研究方向为物联网、边缘计算、机器学习、数据分析等。|朱咏絮(1988-),女,博士,拉夫堡大学研究员,主要研究方向为无线通信性能分析、无线边缘缓存、基于毫米波的无线传输、无人机通信等。|王现斌(1971-),男,博士,西安大略大学电气与计算机工程学院教授、加拿大工程院院士、IEEE院士,主要研究方向为下一代宽带无线移动通信传输理论、5G、通信安全、定位技术、协同通信、异构网融合、边缘计算等关键技术的理论及标准化研究。

Autoencoder neural network-based abnormal data detection in edge computing enabled large-scale IoT systems

Tianqi YU1,Yongxu ZHU2,Xianbin WANG1   

  1. 1 Department of Electrical and Computer Engineering,Western University,London N6A 5B9,Canada
    2 Wolfson School of Mechanical,Electrical and Manufacturing Engineering,Loughborough University,Loughborough LE11 3TU,U.K.
  • Revised:2018-10-29 Online:2018-12-01 Published:2019-01-03

摘要:

物联网以其便于搭建及成本低廉等优点,被广泛地应用于环境监测中。在大规模物联网监测系统中,云平台一直被用作远程的数据和控制中心,然而系统中大量的数据上传和处理给云平台的带宽负载和实时性反馈等方面带来了巨大挑战。因此,提出了基于边缘计算的物联网监测系统框架。边缘计算作为该系统的中间层,能够为终端设备提供实时的本地服务,同时能够通过初步分析分流云平台的计算任务进而降低数据上传量。在此基础上,进一步提出了一种利用自编码神经网络实现的异常检测方法。采用实地采集的海洋气候监测数据进行仿真分析,仿真结果表明,本文提出的基于自编码神经网络的异常检测方法能够充分利用采集数据的空间相关性并准确地检测出异常数据。

关键词: 自编码神经网络, 异常检测, 边缘计算, 物联网

Abstract:

Given the advantages of low cost and easy deployment,large-scale Internet of things (IoT) has been deployed for environment monitoring pervasively.Within such systems,cloud platform is typically utilized as a remote data and control center.However,tremendous amount of data uploading and processing induce huge challenges on bandwidth load and real-time data gathering.In order to overcome these challenges,edge computing enabled IoT system architecture was proposed for environmental monitoring.As the intermediate layer,local processing could be supported for end devices with low latency and assist with preliminary analysis to offload computational tasks from cloud and the amount of data uploading could be reduced.Based on this system architecture,an autoencoder neural network-based abnormal data detection scheme was developed newly.Performance evaluation has been conducted based on the practical oceanic atmospheric data.Simulation results indicate that the proposed scheme can accurately detect the abnormal data by fully exploiting the spatial data correlation.

Key words: autoencoder neural network, abnormal data detection, edge computing, IoT

中图分类号: 

No Suggested Reading articles found!