Digital Twin and Transformer-Based Learning for Secure and Energy-Efficient Active RIS-Assisted Communications

Ensuring secure and energy-efficient connectivity in wireless networks aided by a low-altitude autonomous aerial vehicle (AAV)-mounted active reconfigurable intelligent surface (ARIS) is a major challenge, owing to rapid channel variations, mobility-induced uncertainty, and the heightened risk of adversarial eavesdropping inherent to open-air communication. To address these challenges, we propose a digital twin (DT)-empowered Transformer-based policy learning framework that combines accurate real-time virtual modeling with attention-driven sequential decision-making for AAV–ARIS-assisted communications. The DT enables efficient synchronization with physical channels in dynamic environments, significantly improving data-efficient training. Meanwhile, a Transformer-based actor–critic architecture captures long-range temporal dependencies to jointly optimize ARIS configuration and AAV mobility control across diverse scenarios. A multi-objective reward function is formulated to maximize both secrecy rate and energy efficiency while ensuring robustness against channel uncertainty and adversarial threats. The formulated problem is solved using reinforcement learning with Transformer-enhanced policy optimization. Simulation results demonstrate consistent performance gains over strong baselines, including DT-deep deterministic policy gradient (DDPG), DT-twin delayed DDPG (TD3), and RIS-soft actor-critic (SAC). The proposed framework achieves up to a 12.5% improvement in secrecy rate, a 14.8% enhancement in energy efficiency, and a 33% faster convergence rate. Furthermore, ablation studies and complexity analyses confirm the contribution of each component and demonstrate the scalability of the framework with acceptable latency. These findings highlight the potential of integrating DT technology with Transformer-based reinforcement learning to enable secure, adaptive, and resource-efficient ARIS-assisted communications in next-generation wireless networks.