A Semantic-Driven Data Generative Traffic Modeling Method for Digital Twin Networks

Digital Twin Networks (DTN) create synchronized virtual replicas of physical networks to enable real-time monitoring, prediction, and optimization of network performance, facilitating efficient management in complex environments. However, existing DTN modeling often face challenges in accuracy, efficiency, and high communication costs. This paper proposes a novel semantic-driven traffic generation method that integrates diffusion models, network simulation, and semantic communication for constructing DTN. Our approach first extracts semantic information from real network data and inputs it into Mininet for initial traffic simulation. Then, to reduce discrepancies with real-world data, we introduce a Transformer-based diffusion model, DiffNTG, which refines Mininet-generated traffic to achieve high fidelity. Experimental results demonstrate that our model reduces error metrics between simulated and real networks by up to 96% and decreases computational costs by approximately 83%, showing strong potential for DTN applications.