Inter-slice Resource Dynamic Allocation Algorithm for B5G Services

Abstract

In 5G networks, the network slicing based on SDN and VNF can effectively meet the service requirements of diverse businesses. However, with the increasing number and variety of future network services, existing slicing resource allocation algorithms are difficult to maximize overall network resource efficiency. Therefore, we propose an inter-slices resource allocation algorithm based on Federated Deep Reinforcement Learning (FDRL) for B5G services. We design the user utility function as the weighted sum of spectrum efficiency and different slice user QoS. And set the user utility function and SP cost as optimization objectives. In our algorithm, the global parameters are trained using the FL algorithm framework in the global model. And the local model is trained using the DQN algorithm based on the global parameters and local data. Simulation results show that the proposed algorithm has excellent performance in guaranteeing system utility, spectrum efficiency, and slice users QoS.