•  
  •  
 

Abstract

The Unmanned Aerial Vehicle (UAVs) communication networks are becoming easy targets of cyber-attacks such as denial of service, spoofing and intrusion, compromising on the mission-critical operations. The existing methods for cybersecurity using conventional machine learning or pure deep learning cannot reflect the complexity of spatial and temporal dependencies and cannot be adaptive against new cyber-attacks in UAV networks. In order to surmount these constraints, we present a new Graph Transformer-Temporal Graph Convolutional-Deep Reinforcement Learning (GT-TGC-DRL) architecture to adaptive UAV cyber defense. The methodology models UAV network traffic in terms of graph structures, with nodes used to model UAVs and edges used to model communication links. Graph Transformer module is used to extract spatial communication dependencies, and the Temporal Graph Convolutional Network is used to extract time-related variations of traffic. The extracted spatial-temporal features are given to a Deep Reinforcement Learning agent, which identifies the best defense strategies to counter the identified cyber threats. The framework was implemented on Python and tested with publicly available UAV cyberattack and IDS data. The GT-TGC-DRL model has an accuracy of 98.72%, precision of 98.10%, recall of 97.84%, F1-score of 97.96% and a false positive rate of 1.18 %, which is higher than the previous models such as LightGBM, UAV Hunter, LSTM-GA and cross-layered attention models that had an accuracy of less than 96.34%. This paper illustrates the importance of the proposed hybrid structure in terms of improving resilience by providing intrusion detection and cyber threats mitigation mechanisms.

Download

Share

COinS