Dynamic Routing Method for Integrated Satellite–Terrestrial Multi-Domain Networks Based on Transfer Learning

With the development of communication technology and the diversification of users' needs, the space-earth integrated multi-domain network has become the core of modern communication system. However, rapidly changing topologies, frequent satellite switching, and heterogeneous traffic conditions can challenge classical deep reinforcement learning (DRL) algorithms, affecting real-time accuracy and robust convergence. Given that these networks continue to evolve, rapid adaptation is critical to maintaining data throughput and a seamless user experience. To solve these problems, this paper introduces STGT (Space-Ground Generative Transport), a novel dynamic routing framework that combines generative adversarial network (GAN) with near-end Policy Optimization (PPO) algorithm.