Deep Reinforcement Learning Approach to Slice Admission Control for Resilient 5G Wireless Network with Multidimensional State Space

Abstract

Resource slicing, a critical technique in managing network resources, has been integrated into the latest cellular communication paradigm shift. To enable service providers to set resilient network goals, it is essential to analyze quality of service constrained slice requests using an intelligent algorithm. These detailed characteristics can be modeled as machine learning features and combined with network dynamics to identify hidden network parameters proactively. Slice admission control can be used to examine network dependability and increase network resilience. In this work, we propose a deep deterministic policy gradient technique called sequential twin-critic actor-critic reinforcement learning to evaluate slice requests and perform slice admission control. The proposed approach aims to construct a multistage admission control process for evaluating slice admissibility and adjusting system parameters in a continuous action space. Simulation results demonstrate that our reinforcement learning-based resilience evaluation approach for slice admission control outperforms conventional actor-critic, deep Q-learning, and greedy admission strategies.