AI-Enabled Space-Air-Ground Integrated Networks: Management and Optimization
Peiying Zhang, Ning Chen, Shigen Shen, Shui Yu, Neeraj Kumar, Ching‐Hsien Hsu
Abstract
AI-enabled Beyond 5G (B5G) and 6G technologies are promising candidates to support the future generation Space- Air-Ground Integrated Networks (SAGINs). The highly dynamic heterogeneity and time variability, however, complicate management and optimization efforts. Hence, based on the softwaredefined networking (SDN) technology, in the proposed hierarchical hybrid deep reinforcement learning (HHDRL) method, we demonstrate how one can combine both distributed and central architectures, by deploying local controllers in different domains and global controllers on the whole. It permits us to optimize the network through local fine control and global macro control. We also deploy the DRL models in the controllers, where the optimal policy is learned through the effective interactions between the agent and the environment, as well as via the feedback of the incentive mechanism. Finally, a case study based on resource allocation and related analysis illustrates in detail that the AI algorithm represented by HHDRL will significantly promote the management and optimization process of SAGIN.