Research On Service Migration Algorithm Of Internet Of Vehicles Based On Reinforcement Learning

In the Internet of Vehicles,the cloud computing service carried by the intelligent vehicle terminal will have a long delay.Mobile edge computing can sink the computing power and storage resources of the cloud computing center at the edge of the network,which can effectively solve the delay and energy consumption of mobile services.However,due to the mobility of the vehicle,the vehicle is separated from the coverage of the edge computing node,which interrupts the edge computing service.Therefore,in many delay-sensitive mobile terminal services,the service migration technology in mobile edge computing technology is applied to solve the problem of discontinuous or even interrupted vehicle services.Aiming at the problem of vehicle application service interruption caused by the mobility of intelligent vehicles and the limited communication range of edge nodes in the Internet of Vehicles,this paper focuses on the service migration strategy of vehicle applications when intelligent vehicles move in a certain area,aiming at reducing service delay and communication energy consumption in the edge computing scenarios of single vehicle and multi-vehicle Internet of Vehicles.Firstly,for the single-vehicle application scenario in the Internet of Vehicles,an edge computing service migration strategy based on a reinforcement learning algorithm is proposed.Based on the advantages and disadvantages of the current service migration technology research,a service migration mathematical model of intelligent vehicles in urban roads is established,which comprehensively considers the computing resources of edge computing nodes,network bandwidth capacity,vehicle service delay,and other constraints.Combined with the mathematical model,the onboard task and edge device information are defined as states,the migration decision of virtual machines is defined as actions,and the system utility function jointly defined by delay and energy consumption cost is used as the reward function.Based on the improvement of the Q-Learning algorithm in reinforcement learning,an intelligent optimization algorithm of service migration strategy is obtained.Through algorithm simulation,the optimal service migration strategy of vehicle tasks in a certain time range is obtained.Finally,through comparative experiments,it is found that the PQLA service migration optimization algorithm proposed in this paper can improve the system utility value by about 18 % compared with the MDP algorithm.Then,based on the single-vehicle scenario,the number of intelligent vehicles in the Internet of Vehicles scenario is increased,the mathematical model of multi-vehicle Internet of Vehicles service migration is constructed,and the joint state,joint action,and total reward of multi-vehicle are redefined.Based on the Nash Q-learning algorithm in multi-agent reinforcement learning,the mathematical model is transformed into a mixed-task dynamic random game problem,and the long-term Nash equilibrium solution of multiple intelligent vehicles in the mixed task dynamic random game is solved.Finally,through the simulation comparison experiment of the algorithm,compared with the Q-Learning algorithm of a single vehicle,the NSQL algorithm can achieve a better optimization effect under the premise of ensuring the service quality of vehicle users.