Research On Joint Resource Management Algorithm Of Communication,Caching And Computing For Internet Of Vehicles Based On Deep Reinforcement Learning

Application services in the internet of vehicles(e.g.,infotainment service and real-time high-precision map)often need to meet the requirements of low latency,high reliability or high throughput.In order to meet the requirements of these application services,many technologies have emerged in the internet of vehicles.Among them,the in-network caching caches popular contents to the communication cell,which reducing the transmission of duplicate contents;mobile edge computing performs task offloading closed to the user side,which saving data transmission delay in the core network;the combination of software-defined networking and network function virtualization can achieve network-level resource allocation and flexible task distribution.In order to further improve the service quality of the internet of vehicles,joint research on the above technologies is of great significance.This paper combines software-defined networking and network function virtualization technology to study the joint resource management algorithm of communication,caching and computing for internet of vehicles.The main work is as follows:Firstly,this paper studies the joint resource allocation algorithm of communication,caching and computing for internet of vehicles.For the scenario of vehicles requesting video tasks,this paper establishes the communication model,computing model and caching model,while considering the finiteness of infrastructure resources and the mobility of vehicles,and constructs an optimization problem to maximize the throughput of vehicles requesting video task.According to the optimization problem,a deep Q network(DQN)algorithm is proposed.In the beginning,the physical quantity defined in the model is used as the input state,and the connection between the vehicle and the infrastructure is used as the output action.Then the reward function considers the greedy strategy and overload penalty,and the neural network structure is designed and used for DQN algorithm training finally.The simulation analyzes the influence of the number of vehicles and the average content request size on the system throughput.When the throughput is saturated,the throughput of the proposed algorithm is about15% higher than that of the greedy algorithm,indicating that the proposed algorithm has better system throughput.Secondly,this paper studies the joint resource allocation algorithm of communication and computing for internet of vehicles.For the scenario of vehicles requesting real-time highprecision map,this paper establishes the communication model and computing model.Based on the established model and load balancing evaluation index,the optimization problem of load balancing for vehicle communication and computational offloading is constructed under consideration of communication and computing resource constraints and vehicle mobility constraints.According to the optimization problem,the input state,output action,reward function and neural network structure of DQN algorithm are designed respectively.The simulation compares the performance of different optimization schemes in load balancing of internet of vehicles.In terms of the variance of downlink rate and computing rate,the variance of the greedy algorithm is about 4 times that of the proposed algorithm,indicating that the proposed algorithm has better system fairness.