Cross-domain data-driven reinforcement learning for IGSO satellite coverage optimization

The beam coverage area of Inclined GeoSynchronous Orbit (IGSO) satellites suffers from a diurnal north-south shift caused by the orbital inclination. Thus, it is crucial to adjust the satellite's attitude for stabilizing the beam coverage area and enhancing the communication signal quality. However, traditional optimization methods fail to optimize attitude due to the intricate orbital environment and the time-consuming simulation solving process. In this paper, we propose a Cross-Domain Data-Driven Reinforcement Learning (C3DRL) algorithm for IGSO satellite coverage optimization, which utilizes offline expert data to assist agent learning and exploration during interactions within the online environment. To overcome the dynamic differences between online and offline environments, a practical cross-domain adaptation approach is proposed. This approach distinguishes dynamics changes and penalizes agents for learning transitions with significant dynamics differences. By addressing the issue of action value function overestimation, we effectively utilize offline data to drive agent learning process. The experimental results show that C3DRL effectively stabilizes the coverage area compared with other algorithms. In addition, several ablation experiments are conducted to demonstrate the effectiveness of the cross-domain adaptation method.