Edge intelligence drives multiscale anomaly detection network for heterogeneous IoT traffic

Abstract

When edge nodes undertake the task of near-source traffic monitoring in the Internet of Things (iot), limited computing resources and the difference of abnormal behavior time span are easy to cause local mutation and insufficient information representation of continuous attack. To solve this problem, this paper constructs a multi-scale anomaly detection network for heterogeneous traffic driven by edge intelligence. The traffic information is extracted at the edge side, and the communication changes of different persistence ranges are depicted through short, medium and long time scale coding, and the dynamic fusion is performed according to the scale contribution to realize the identification of normal communication and multi-class abnormal traffic. Experiments were carried out based on the CICIoT2023 dataset. The detection accuracy of DDoS, DoS, Recon, Web, Brute Force, Spoofing and Mirai traffic in our network reaches 99.2%, 98.9%, 97.8%, 96.4%, 96.9%, 98.1% and 99.0%, respectively. The accuracy is 97.88% when the traffic scale is 90000, and the average detection delay is 18.6 ms when the proportion of available computing resources of edge nodes is 20%. The results show that the constructed network can balance the accuracy of heterogeneous traffic anomaly detection, the stability of scale change and the real-time performance of edge deployment, which can provide technical support for the edge side anomaly communication monitoring of iot.