Deep Reinforced Adaptive Graph Optimization Network for Intelligent Intrusion Detection in UAV Communication Networks

Abstract

Unmanned Aerial Vehicle (UAV) communication networks integrated with Software Defined Networking (SDN), Internet of Things (IoT), and 5G technologies are increasingly vulnerable to cyberattacks such as spoofing, flooding, routing manipulation, and abnormal communication intrusions. Existing intrusion detection systems suffer from limitations including high computational complexity, increased false positive rate, scalability issues, and poor adaptability to dynamic UAV topologies. To address these challenges, this research proposes a novel Deep Reinforced Adaptive Graph Optimization Network (DRAGrON) for intelligent intrusion detection in UAV communication environments. The main objective of the proposed framework is to improve real-time attack detection accuracy while reducing detection delay, energy consumption, and communication overhead. The DRAGrON framework integrates Adaptive Graph Optimization, Graph Neural Networks (GNN), Attention Mechanisms, and Deep Reinforcement Learning (DRL) to analyze UAV communication relationships and dynamically optimize intrusion detection policies. The experimental validation under multiple intrusion scenarios that are demonstrated that the proposed DRAGrON framework achieved superior performance with 98.7% accuracy, 98.1% precision, 97.8% recall, 97.9% F1-score, 1.8% false positive rate, and 0.31 s detection delay. Additionally, the framework achieved 96 Mbps throughput, 98.8% packet delivery ratio, 0.51 J energy consumption, and 1320 rounds network lifetime. The results confirm that DRAGrON provides efficient, scalable, and real-time intrusion detection for secure UAV communication systems.