Research On Network Slicing Resource Allocation Algorithm Based On Deep Reinforcement Learning

The rapid growth of user access equipment and network traffic brings tremendous pressure to operators.In order to meet the diverse demands of users,Network Slicing technology emerges.Compared with traditional networks,which could only satisfy a single service demand,Network Slicing allows the establishment of multiple logical sub-networks in a physical network to meet various services' demands.However,resource allocation and management among these multiple sub-logical networks is a challenging problem in Network Slicing.This paper proposes a network slicing resource allocation algorithm based on Deep Reinforcement Learning(DRL)to improve network slicing communication quality and resource utilization.In order to speed up the convergence of the model,we introduce the idea of transfer learning into the A2 C algorithm,and propose a reward function that considers progress and retreat.The main contributions of this paper are as follows:1)In the scenario of a single base station and multiple slices,considering the dynamics of network requests,we employ the network slice resource allocation to maximize system utility,meanwhile,improve system resource utilization and satisfaction with slices.2)We formulate the strategy of network slicing resource allocation as a Markov process and apply the DRL algorithm to solve resource allocation.Then,based on the A2 C algorithm,we build a resource allocation model for the network slicing scenario,which can implement reasonable resource allocation actions to meet the status of network,thereby improving system utility.3)To accelerate the convergence in model training,we propose two optimization schemes in this paper.The first is to optimize the reward function and put forward the concept of relative reward based on progress and retreat;the second is to design the transfer A2 C algorithm,which introduces the prior knowledge of the high reward state-action relationship into the AC network,thereby accelerating the learning speed.We implement the proposed network slicing resource allocation algorithm,and simulate the models with contrastive analysis.Through the experimental simulation results,it can be concluded that the resource allocation strategy algorithm based on DRL not only guarantees user quality of service but also improves resource utilization in dynamic networks.Besides,the two model learning optimization methods proposed in this paper can also effectively increase the learning speed and accelerate the training convergence process.