On self-adaptive 5G network slice QoS management system: a deep reinforcement learning approach

Along with the development of mobile network communication standards to the fifth generation, the complexity of network usage patterns has increased. The concept of network slicing is proposed to improve the utilization rate of network and computing resources, and to provide corresponding service quality for different network applications. In this paper, we propose a self-adaptive quality of service (QoS) management system which can be added to the 5G core network architecture, using network usage behavior and service level agreements (SLA) to generate corresponding QoS marking rules and enhance 5G core networks' QoS mechanism. In response to the fact that user behavior changes over time, our system leverages deep reinforcement learning methods to dynamically generate QoS marking rules based on user behavior. In terms of experiments, we use a NS-3 network simulator to initially validate the system and observe that, as the training progresses, the measured network QoS KPIs of users become closer to the SLA.