Dynamic Resource Planning Based On Deep Reinforcement Learning In Vehicular Edge Computing Networks

In recent years,with the development of Internet of Vehicles technology and continuous update of on-board equipment,a large number of computation-intensive and delay-sensitive vehicular applications have appeared,which put forward strict requirements on the computing power,bandwidth and data processing capacity of terminals.The vehicular edge computing system formed by applying mobile edge computing technology vehicular network can provide computing services for other mobile devices through task offloading.This computing paradigm can improve the service quality of vehicular devices and the utilization rate of resources in the system.However,due to the mobility of vehicles,the vehicular task offloading environment is dynamic and uncertain,with rapidly changing network topology,wireless channel state and computing load.These uncertainties make the task unloading process non-idealized.In addition,traditional V2V-based task offloading schemes pay more attention to resource allocation,seldom consider communication security,and ignore the continuous motion space during task uninstallation.To solve these problems,this paper proposes an efficient V2 V offloading solution managed by the RSU,which improves the utilization of idle computing resources on vehicle-mounted devices and reduces task latency by optimizing offloading decisions and resource allocation,improving system security and satisfying user experience quality.The main research contents are as follows:1.In the vehicular edge computing system,in view of the uncertainty of the vehicles task offloading environment,this paper sinking the computing resources of the MEC server to the vehicles and studying the solution of V2V-based task offloading,so that the vehicle can learn the service performance of the surrounding vehicles and offload the task without the unknown state information.Based on the multi-armed bandits framework,a reinforcement learning algorithm of second-order exploration is designed.The algorithm takes the second-order unloading reward difference into the exploration item,further improves the confidence space suitable for the decision object in the scenario of unknown state information,so as to maximize the average unloading reward of users.In addition,a service set update method is proposed after the end of one offloading stage.To ensure the quality of service for users.Simulation results show that,compared with the existing algorithm based on confidence upper limit,the offloading reward under the proposed scheme is improved by about 34%.2.To solve the communication security problem in VEC,an efficient credit-based V2 V collaborative task offloading scheme is proposed.The scheme considers the potential data security problems caused by task transmission failure or long computing delay of resource providers,makes reliable offloading decisions for vehicles,and guides communication and computing resource allocation.Firstly,a reputation mechanism is designed to evaluate vehicle service performance during modeling,which takes resource allocation,task transmission success rate,processing delay and other factors into consideration.Secondly,a task offloading algorithm based on Deep Deterministic Policy Gradient(DDPG)is designed to obtain the optimal offloading strategy in the collaborative environment of vehicles,effectively solve the continuous control problem,and achieve fast convergence.Finally,a lot of simulation experiments are carried out,and the results show that the performance of this scheme is improved by 16.3% on average compared with some existing schemes.