Offloading Algorithm for Multi-Agent and Double-Layer Offloading in Internet of Vehicle
In the edge computing environment of the Internet of Vehicles(IoV),with the aim of efficiently offloading tasks and allocating resources to alleviate the limited storage and computing power of vehicles,this study proposes a offloading algorithm based on multi-agent and double-layer offloading in IoV.A three-layer network model consisting of a Mobile Edge Computing(MEC)server,Vehicles,and Non-Task Vehicle Cloud(MEC-V-NTVC)interconnection is first proposed.Additionally,task,judgment,and calculation models are established.Second,the computational offloading and resource allocation of task vehicles are abstracted into a Partially Observable Markov Decision Process(POMDP),and a double-layer offloading mechanism is proposed to minimize the system cost.Applying a double-layer offloading mechanism and monotonic value function factorization for deep multi-agent reinforcement learning QMIX,a deep reinforcement learning offloading algorithm DLSQMIX based on the double-layer offloading mechanism is proposed,which coordinates task vehicles,non-task vehicles,and state information,considering the time constraint and cooperates with the computation power of the MEC and the non-task vehicle cloud,to learn optimal offloading decisions.The system overhead and latency are compared and explained in terms of the computing power of edge servers and non-task vehicles,number of task vehicles and non-task vehicles,and average task volume.The simulation experiment results demonstrate that the DLSQMIX algorithm can effectively solve the task-offloading problem.Compared with the Genetic Algorithm(GA),Particle Swarm Optimization(PSO)algorithm,and QMIX algorithm,the proposed algorithm reduces the system overhead by 2.52％-3.91％and latency by 3.50％-6.59％.
Internet of Vehicle(IoV)edge computingnon-task vehicle clouddouble-layer offloading mechanismmonotonic value function factorization
万方数据
维普
