Enhancing Edge Intelligence in Wireless Communication Networks Using Large Models for Security and Adaptive Control

Abstract

Wireless communication networks (WCN) are becoming more complicated and dynamic, especially when it comes to edge computing. As a result, intelligent, self-governing systems that can manage security and control duties in real time are required. This paper presents a novel Edge Transformer for Security and Adaptive Control (EdgeFormer-SAC), a Transformer-based large model (LM) designed for edge environments that is compact and effective. Using a compressed Transformer architecture designed for low-latency and low-energy situations, the novel EdgeFormer-SAC combines security anomaly detection and adaptive control to jointly manage multi-task learning at the wireless edge. The proposed EdgeFormer-SAC model has been evaluated against well-known machine learning (ML) models including Support Vector Machine (SVM), deep learning (DL) models including Long-Short Term Memory (LSTM), Mobile Network Version 2 (MobileNetV2), Tiny Bidirectional Encoder Representations from Transformers (TinyBERT), and Deep Reinforcement Learning Agent (DRL) techniques through extensive tests in simulated wireless environments. The proposed EdgeFormer-SAC model maintained a real-time latency of 17.5 ms and low energy consumption at 1.3 W, while achieving the greatest accuracy and F1-score of 94.8% and 93%, respectively, and a false positive rate (FPR) of only 2.3% and an adaptation score of 89%.