Research On The Performance Of Vehicular Networks Based On Deep Reinforcement Learning

Vehicular networks are new communication networks that use vehicles as infor-mation sensing objects to realize the interconnection between vehicles and people,machines and things,and are the basis for driverless cars,smart cars and intelligent transportation,which are receiving wide attention from academia and industry.Vehic-ular networks communication performance is the key factor that restricts its deploy-ment and application,especially the communication link rate fairness performance as well as security performance.The rate fairness performance aims to avoid the uneven allocation of resources such as spectrum and transmission power,which leads to the low rate of some communication links and cannot meet the basic needs of normal ve-hicular communication.The security performance aims to ensure the security of in-formation transmission process and avoid malicious eavesdropping of users’private information.Therefore,this thesis improves the communication link rate fairness per-formance and security performance by optimizing the allocation of spectrum and transmission power.The main research work is outlined in three parts as follows.(1)Two typical vehicular communication scenarios are constructed:one scenario consists of a base station(BS),multiple V2V(Vehicle-to-Vehicle)users,and multiple V2I(Vehicle-to-Infrastructure)users;the other scenario consists of a BS,multiple V2V users,multiple V2I users,and a malicious eavesdropper E.At the same time,a channel model of the communication link is developed,as well as link constraints are designed for the diverse quality of service requirements of vehicular networks.(2)Based on the first scenario above,the rate fairness performance of V2I com-munication links in vehicular networks is studied.Firstly,a rate fairness indexF_c is designed to measure the rate fairness of V2I links,and then an optimization problem with the objective of maximizing rate fairness is established.This is a nonlinear and nonconvex optimization problem,which is usually difficult to solve.Therefore,a deep reinforcement learning(DRL)algorithm is proposed for solving this optimization problem.The algorithm can optimize the allocation of spectrum and power resources to maximize the fairness indexF_c,and thus improve the fairness performance of communication rates.Simulation experiments show that the proposed DRL algorithm can effectively improve the rate fairness performance of V2I communication links in vehicular networks.(3)Based on the second scenario above,we study the rate security performance of V2V communication links in vehicular networks.We first model the rate performance,then establish the optimization problem to maximize the rate security performance,and finally propose a DRL algorithm to solve the optimization problem.Under this algo-rithm,the rate security performance is improved by optimizing the allocation of spec-trum and power resources.Simulations show that the DRL algorithm proposed in this thesis can effectively improve the security performance of V2V communication link rate in vehicular networks.Figure 28 Table 7 Reference 85...