Research On Resource Allocation Algorithm Of Wireless Communication Network Based On Reinforcement Learning

With the rapid development of communication technology and smart terminals,various service needs are growing rapidly,which makes the data traffic carried by cellular system grow explosively.Traditional network architectures cannot encounter the everincreasing demand for mobile service,meanwhile,the battery storage and computing capabilities of terminal equipment cannot encounter the needs of a great number of service requirements.In recent years,the MEC technology transfers computing tasks to the edge of the network,which can extend battery life and provide reliable data processing and storage capabilities;In addition,D2D technology can achieve direct communication between terminals,thereby reducing the load on the cellular network.Moreover,the relayaided D2D communication can effectively solve the problem of long-distance between D2D users,expand the radius and improve the transmission rate of D2D communication.The main research content includes the following two parts:1)Under the network scenario of multiple devices and a single MEC server,a dynamic offloading decision of application tasks and the allocation of communication resources are studied.Under the conditions of the computing task queue's stability and time delay limitation,and maximum power constraints of the mobile devices,an optimization model is established to achieve task offloading decision-making,computing resource,wireless channel and transmission power allocation in order to minimize the system's long-term average energy consumption.In view of the limited battery storage and computing capabilities of terminal equipment,the dynamic nature of the MEC network environment,the multi-optimization parameters and the interrelated characteristics,the model is simplified to a joint optimization problem of channel and power allocation,and an optimization plan that simplifies parameters and reduces computational complexity is proposed.Then,the deep Q-network(DQN)method was used to realize a resource allocation algorithm for long-term stability of the computing task queue under the constraints of power and delay.The proposed algorithm proved by simulation can effectively promote the energy efficiency and improve the data processing ratio of system.2)In view of the long distance and poor channel conditions between D2D users,which lead to the decline of communication quality and cannot meet the service requirements,and there is co-channel interference between D2D users and cellular users,the resource allocation problem in relay-aided D2D communication underlay cellular networks is studied.The optimal resource allocation algorithm for mode selection,relay selection,and channel allocation in relay-aided D2D communication with lower computational complexity is proposed based on the existed power allocation scheme.Firstly,the joint channel allocation,mode selection and relay selection scheme of relayaided D2D communication is modeled as a finite MDP,and a deep MCTS model is established.The mode is composed of deep residual network and MCTS.MCTS uses the deep residual network to generate the priors of actions probability and action value to evaluate the choice of execution action.Secondly,the deep residual network uses the optimal value obtained in the MCTS search process as a label for training and updating the deep residual network parameters.Finally,an intelligent resource allocation strategy can be obtained under QoS constraints.The simulation results show that under the same conditions,the proposed algorithm is significantly higher than the transmission rate obtained by the linear programming algorithm.