Task Offloading Algorithm Based On Edge Intelligence In Vehicular Communications

The Internet of Vehicles is one of the representative application scenarios of 5G mobile communication.With the development of information technology and new business requirements,a large number of vehicular applications with high demand for computing and storage resources have emerged.However,in the current vehicle communication environment,the traditional task migration and resource allocation algorithms are difficult to meet the requirements of low delay and high reliability of new vehicular applications such as real-time road conditions and intelligent recognition.Mobile Edge Computing(MEC)can deploy a large number of high-performance servers at the network edge of new application scenarios such as intelligent transportation.The tasks of vehicles can be offloaded to MEC edge server for execution,which can effectively reduce the task execution delay.Therefore,two different dynamic task offloading algorithms based on Deep Reinforcement Learning(DRL)are proposed for different application scenarios.(1)For the single-cell scenarios with multiple vehicles,a hierarchical task offloading algorithm based on Deep Q Netwok(DQN)is proposed,which makes the vehicle maximize the processing speed of tasks under the given energy budget.Firstly,the Analytic Hierarchy Process(AHP)is introduced to divide the priority of the computing intensive tasks carried by the vehicles,and the system model is established with the weighted sum of the processing rate of the tasks as the optimization objective.Secondly,an AHP-DQN task hierarchical offloading algorithm is proposed,which enables the vehicles to make the optimal task offloading strategy adaptively in the time-varying wireless channel environment.Simulation results show that compared with DQN algorithm,the task execution delay of the proposed algorithm is reduced by 14%,the success rate of task offloading is increased by 11%,and the task execution efficiency of vehicles is effectively improved.(2)For the multi-cell with multiple vehicles scenario,considering the challenge of the lack of MEC servers on the edge side,a joint optimization algorithm of dynamic task offloading and resource scheduling is proposed.Firstly,based on the queuing theory,the system model is established by taking the sum of the execution delay of all tasks in the vehicular edge network as the optimization objective;then the over estimation problem in DQN is discussed,and the impact of over estimation problem is alleviated by the Double DQN algorithm.Finally,a Double DQN based algorithm for computing resource allocation and dynamic task offloading of vehicles is proposed.Simulation results show that the proposed algorithm can greatly improve the utilization of edge computing and storage resources,effectively reduce the task execution delay of vehicles,and the convergence speed of the algorithm is higher than the DQN algorithm.