Reinforcement Learning Assisted Bandwidth Aware Virtual Network Resource Allocation

Space-air-ground integration to support seamless coverage of ground, satellite, airborne, and marine communications, is likely to be a key trend in the 6G era. One of several key challenges in such space-air-ground integration networks (SAGINs) is to design efficient scheduling approaches for multi-dimension network resources. Due to the inherent heterogeneity characteristics, we demonstrate how can transform the network resource allocation problem in SAGINs into a multi-domain virtual network resource allocation problem, as well as proposing a reinforcement learning assisted bandwidth aware virtual network resource allocation algorithm (RL-BA-VNA). Specifically, RL-BA-VNA leverages reinforcement learning and uses a policy network as an agent to perform the node embedding. In order to support users' exacting bandwidth requirements, we prefer to select virtual network requests with large bandwidth for embedding. Experiment findings show that the proposed algorithm RL-BA-VNA outperforms respectively the other three conventional virtual network resource allocation algorithms RL, DRL and BASELINE by an average of 2.06%, 4.93%, 11.07% in terms of long-term average reward, acceptance rate, and long term reward/cost.