Research And Realization Of Resource Optimization In Internet Of Vehicles Based On Swarm Intelligence

With the development of technologies such as automatic driving and vehicle-road collaboration,intelligent networked vehicles need more roadside edge computing capabilities for environmental awareness and situational awareness to ensure driving safety,improve traffic efficiency,and meet richer Internet of Vehicle services demand.Due to the large-scale fast-moving characteristics of vehicles,intelligent vehicles will face the migration of Internet of Vehicle services and the offloading of vehicle computing tasks during the use of Internet of Vehicle resources,which will lead to contention for the communication resources and edge computing resources of Internet of Vehicle.Due to the mobility of the vehicle,Internet of Vehicle services need to migrate between edge computing units to enable the vehicle and the edge computing unit providing the service to transmit data with a low communication delay.Due to the limited resources of edge computing units,the consumption of edge computing unit resources needs to be considered during service migration.At the same time,there is overhead when services are migrated.Aiming at the above service migration problems,a service migration strategy based on Multi-Agent Reinforcement Learning is proposed in the thesis.This strategy optimizes Internet of Vehicle resources by selecting appropriate service migration objects for vehicles,and based on the distributed execution of multiple agents,it quickly provides service migration decisions for vehicles.Considering the vehicle demand for task offloading of the edge computing capabilities,it is necessary to avoid task processing timeout due to the high load of the edge computing unit.A strategy for vehicle task offloading based on discrete particle swarm optimization is proposed in the thesis.This strategy can reduce the network overhead while ensuring task delay requirements.Finally,experiments are performed on the service migration strategy based on Multi-Agent Reinforcement Learning and the task unloading strategy based on discrete particle swarm optimization.Experimental results validate the efficiency of the proposed methods.