Research On Task Offloading Algorithm For Vehicle Edge Computing Oriented To Latency Optimization

Mobile Edge Computing(MEC)is a distributed computing architecture that can transfer the computing power from the cloud center to the network edge.It has the characteristics of a wide range of applicable scenarios and fast response speed.In scenarios such as highways,mobile devices represented by vehicles generate a large number of computing tasks.These tasks have different characteristic attributes such as data size,delay requirements,and so on.In addition,the high-speed movement of vehicles can cause network links to switch frequently.In this situation,the computing power of the vehicles themselves cannot meet the high requirements of computation-intensive and delay-sensitive tasks for computing power and delay.Therefore,it is necessary to rely on the Mobile Edge Computing architecture to offload these computing tasks.The main work and contributions completed in this article include the following aspects:(1)In order to solve the problems of low success rate and high delay of task unloading caused by the mobility of mobile terminal devices and system load status in the process of traditional mobile edge computing task unloading,this paper proposes a task unloading algorithm based on deep reinforcement learning.The algorithm first obtains the environment information including vehicle location,vehicle speed,task attribute information,network load and server load.Secondly,it defines the Quality of Experience(QoE)formula to consider the service time and task loss of tasks to solve the problem that the offloading decision will sacrifice a large number of other tasks to meet the service time limit of some individual tasks,which leads to the performance degradation of the overall task offloading system when only the task service time is used as the judgment criterion.Then,the reward functions including QoE,network load and server load are defined,and the coefficients of the above three parameters can be shifted to optimize the performance of the corresponding three directions according to different training objectives in order to improve the environmental adaptability of the algorithm.The experimental results show that the algorithm has good performance on task average service time,task average failure rate,task average QoE and other indicators.(2)When the number of tasks is large,the edge server does not have enough storage space to receive and process tasks,and a large number of tasks are centrally offloaded to the cloud server,resulting in network congestion,which in turn leads to a large number of task execution failures.The caching algorithm is introduced to improve the storage performance of the edge server and reduce the failure rate due to insufficient storage space of the edge server.First,a task popularity model is proposed based on Newton’s cooling law,and a deep Q-network-based algorithm is used to make decisions about the cache contents of the edge server and improve the cache efficiency.Experiments are designed to compare the deep Qnetwork-based caching algorithm with traditional caching algorithms,and the experimental results show that the DQN-based caching algorithm outperforms the other compared algorithms.Second,the task offloading algorithm is combined with the deep Q-network-based caching algorithm to solve the bottleneck limitation of insufficient storage space of the edge server in the high-load state and improve the system performance.The experiments are designed to compare the joint offloading and caching algorithm with the separate offloading algorithm,and the experimental results show that the performance of the offloading system with the caching algorithm is significantly improved.