Task-Oriented Service Provision and Scheduling for Non-Terrestrial Networks: A Deep Reinforcement Learning Based Approach

The non-terrestrial network (NTN) receives considerable attentions due to the development of the sixth generation technologies in recent years. However, the mobility and heterogeneity of satellites and unmanned aerial vehicles in NTN lead to increasing challenges in resource allocation and management. As such, the network function virtualization technology is introduced, by deploying service function chains (SFCs) to achieve requirements. However, it brings conflicts and resource competition due to urgent unexpected tasks. Therefore, scheduling multiple SFCs in NTN is a crucial issue. In this paper, a task-oriented service provision and scheduling model is proposed to tackle the dynamic complexity of NTN by designing the reconfigurable time extension graph. Subsequently, the SFC scheduling problem is formulated to maximize the number of completed tasks within limited resources and time constraints. Due to the problem form of integer linear programming, we propose a deep reinforcement learning based algorithm. Finally, simulation results validate the effectiveness of the proposed algorithm, indicating its superior performance over other benchmark algorithms.