Digital Twin-assisted AI-driven Security Framework for 6G LEO Satellite Networks

Low Earth Orbit (LEO) satellite networks are becoming a critical component of next-generation communication infrastructures. However, as these networks expand, they face increasing cybersecurity threats, which can severely disrupt inter-satellite communications and lessen network reliability. Traditional anomaly detection methods often struggle with high false positive rates and scalability challenges, making them insufficient for securing large-scale satellite constellations. To address these challenges, we propose a digital twin-assisted adaptive security framework, integrating Long Short-Term Memory (LSTM) networks for sequential anomaly detection and Deep Q-Networks (DQN) for dynamic adaptive mitigation. Additionally, we introduce an automated neighbour selection mechanism for federated learning, enabling satellites to collaboratively update security models while minimizing communication overhead. Simulation results demonstrate that our framework achieves 20.4% higher anomaly detection efficiency compared to One-Class SVM (OC-SVM) under increasing attack intensity. Furthermore, our automated neighbour selection strategy reduces model synchronization overhead by 58.49%, ensuring scalable and efficient model updates. While network delay increases as more satellites participate in federated learning, our approach keeps it within an operationally acceptable range, balancing real-time detection accuracy and communication efficiency. By integrating digital twins and and AI-driven approach, our framework ensures scalable, adaptive, and resilient security for 6G-enabled LEO satellite networks, effectively countering evolving cyberattacks while preserving real-time network stability and operational efficiency.