Research On Wireless Virtual Network Resource Allocation Based On Deep Reinforcement Learning

In the future 5G network,due to the diversity of services,network slicing technology is a very important technology,which can be dynamically configured according to the service needs of service providers.Compared to traditional physical networks,network slicing is more flexible and can also reduce costs to some extent.However,since slices share the wireless resources of the entire system,the allocation and isolation of resources between slices is a key problem.The network slice can be performed on both the Radio Access Network(RAN)and the core network,but the network slice on the RAN side is discussed in this paper.Considering the dynamic nature of the network request,the load between the slices will also change,which will result in excess or insufficient resources of the slice.Therefore,this paper builds a wireless virtual network resource management system based on the architecture of software-defined network(SDN).The main purpose is to maximize the resource utilization of the entire system under the premise of ensuring quality of service(QoS)satisfaction.In order to dynamically adapt to changes in network requirements,this paper proposes two resource reservation schemes based on deep reinforcement learning(DRL)algorithm framework: one is based on deep Q-network(DQN)resource reservation scheme,and the other is resource reservation scheme based on dueling deep Q-network(Dueling DQN).Therefore,the main work of this paper is:(1)Resource reservation strategy based on DQN.DQN is a deep reinforcement learning framework that continuously interacts with the environment to optimize network parameters,so it can adapt to dynamic network requirements.For a network scenario with multiple heterogeneous slices,all slices share the resources of the entire system,and unreasonable reservations may result in excess or insufficient resources of the slices.Therefore,this paper proposes a resource reservation strategy based on DQN,which can make reasonable resource adjustment actions according to the current environmental state,thereby improving the resource utilization of the system.(2)Resource reservation strategy based on Dueling DQN.In order to better balance the user satisfaction and resource utilization of the slice,this paper proposes a resource reservation strategy based on Dueling DQN to dynamically adjust the proportion of resources between slices.In terms of physical resource allocation,this paper proposes a shape-based allocation scheme to model the problem as a two-dimensional knapsack problem.A less complex heuristic algorithm is used to solve the problem,and as many data flows are scheduled as possible on a limited spectrum resource.Thereby,the idle resource block will be less and the resource utilization will be higher.Based on the above research content,some simulation experiments are carried out to evaluate the performance of DQN-based resource reservation strategy and Dueling DQN-based resource reservation strategy.We also compared some existing methods to evaluate the performance of the two methods presented in this paper.