Long-short term anomaly detection in wireless sensor networks based on spatio-temporal correlation in IoT systems
Zhihao Wei, M. Li, Jianghao Wei · 2023
For the large number of different information sources collected by heterogeneous wireless sensor networks, traditional anomaly detection frameworks usually focus on the data analysis itself and lack a focus on unexpected sensor data, leading to unnecessary waste of resources. To address this problem, this paper proposes a method for automatic anomaly detection in heterogeneous sensor networks that combines edge data analysis and cloud data analysis. The former utilizes a completely unsupervised Isolation forest algorithm, while the cloud data analysis utilizes a Long-short term memory neural network algorithm. Experimental evaluation is performed by discussing the analysis of the proposed method using discrete and event outlier data. The experimental results show that the proposed method can reasonably cope with discrete outliers and event outliers, and the detection accuracy of event outliers is improved from 93.60% to 98.91%. At the same time, the Edge-Cloud combination alleviates to a certain extent the drawback of high energy consumption and latency of the traditional machine learning algorithm local combination model.