Research On Task Offloading Strategy Of Internet Of Vehicles Based On Mobile Edge Computing

With the development of 5G wireless communication networks and the Internet of Vehicles,a new generation of smart terminal equipment generates a large amount of data,which has higher requirements for computing and communication security and service quality.Traditional cloud computing can no longer meet the needs of computing-intensive users with low latency and high reliability requirements.In the current vehicular communication environment,where the amount of computational data generated by vehicular tasks within strict processing deadlines is growing exponentially,communication and computational resource constraints remain a bottleneck.How to meet the computing needs of new vehicle applications such as real-time road conditions and intelligent recognition is a key challenge.In recent years,due to mobile edge computing,it can sink to the vicinity of mobile users to provide offloading services.Therefore,a large number of high-performance servers can be deployed at the network edge of intelligent transportation,reducing the time delay of new vehicle applications.However,different offloading strategies have a large impact on latency and communication computation cost.When communicating with highspeed moving vehicles in the Internet of Vehicles,the return delay between the user vehicle and the downlink transmission of edge device is easily ignored,which makes delay-sensitive tasks unable to meet the delay requirements.On the other hand,considering that there are a large number of delay-sensitive tasks in the city due to special events,planning intelligent rental service vehicle(RSV)travel path matching task nodes to provide offloading services for it.There will be problems such as uneven distribution of vehicle locations,untimely response to task requirements,unreasonable task processing rate,and high cost.Therefore,it is necessary to rationally allocate RSV and plan routes according to user needs,provide computing services,and meet user real-time needs.In response to the above problems,this thesis focuses on the vehicle edge computing network,and proposes an offloading strategy based on reinforcement learning and a path planning and task offloading matching strategy for intelligent RSVs based on genetic algorithms for different usage scenarios.The main work of this thesis is as follows:(1)Aiming at the task scenario where the mobility of moving vehicles leads to the back offload return delay problem of downlink transmission,this thesis studies the task offloading strategy to provide stable computing,communication and storage services for user vehicles in the Internet of Vehicles.In this scenario,the offloading problem is formulated to minimize the delay cost and communication computation cost under the maximum delay constraint.In this thesis,the Vehicle Adaptive Feedback(VAF)algorithm is designed considering the position,speed and computing resources of the vehicles to obtain the priority order of potential service vehicles and obtain the set of service vehicles.Then,combined with the state of the base station,roadside units and vehicles in the service vehicle set,the problem is planned as a Markov decision process,and the optimal offloading scheme is obtained by using the proposed Vehicle Deep Q-network(VDQN)algorithm.In particular,the problem of interruption due to vehicle movement is formulated as a return function to evaluate the task offloading strategy.The simulation results show that the proposed scheme significantly improves the quality of service(Qo S).(2)Aiming at the delay-sensitive task scenarios in urban areas due to the surge of special events,this thesis studies the route optimization and task scheduling problems of RSVs in the system,and balances the relationship between delay and expenditure.The problem is formulated as minimizing offloading delays and costs under the constraints of ensuring passenger satisfaction.This thesis proposes a joint scheduling and path optimization non-dominated sorting genetic algorithm(JSPO-NSGA)to optimize system performance.First,the accessibility graph is designed for the task nodes,and the task scheduling matrix is initialized based on the shortest travel time algorithm(STTA)and the optimal scheduling algorithm(OSA)based on travel time weights and RSV selection probability.Then,the JSPO-NSGA is designed to generate multiple Pareto optimal solutions to the RSV routing optimization and task scheduling problems.Finally,a weighted sum method is used to find the recommended solution.In particular,the JSPO-NSGA establishes a mapping relationship for the effective genes of crossover individuals,and maps the invalid solutions generated by crossover to effective solutions through conflict detection.Simulation results show that the proposed scheme has superior performance.