Open AccessAnomaly Detection for Internet of Things Based on Compressed Sensing and Online Extreme Learning Machine Autoencoder
Author(s) -
Yun Yu,
Xiaojun Wu,
Sheng Yuan
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1544/1/012027
Subject(s) - autoencoder , computer science , anomaly detection , wireless sensor network , machine learning , artificial intelligence , internet of things , real time computing , enhanced data rates for gsm evolution , task (project management) , data mining , deep learning , computer network , embedded system , management , economics
Abnormal events refer to specific events, such as forest fire, occurring in the wireless sensor networks of the Internet of Things (IoT), whose behaviors are quite different from normal events. By learning the underlying structure of sensor data, users can be helped to learn about the occurrence of these events as soon as possible. Due to the large number of sensors in the IoT and the periodic collections of data, sensor data has the problems of high dimensions and large amount, and the transmission of a large amount of data in the network is not a small challenge for bandwidth. In addition, the sensor data is unlabeled, so it is time-consuming and unrealistic to manually label all the data. Abnormal events in the IoT require low delay, such as gas concentration monitoring, In the IoT, data is generated continuously, so a well-trained model cannot remain unchanged, and features of new data need to be continuously learned. On account of the limitation of hardware, edge nodes cannot undertake the complicated and time-consuming task of model training and detection. None of the existing algorithms can meet the above requirements well, so this paper proposed an algorithm based on Compressed Sensing and Online Extreme Learning Machine Autoencoder named COELMAE. The proposed algorithm can carry out anomaly detection in low delay, unsupervised and online learning, and also has low computational complexity. What’s more, the algorithm can reduce the amount of data transferred about 60%.