The Research On Multi-access Edge Computing Offloading And Framed Slot Allocation Scheme In The Internet Of Vehicles

The Multi-access Edge Computing(MEC)-enabled vehicular networks have gained extensive attention from academic and industrial community.A large number of research focus on selecting offloading nodes and communication resources allocation,which is a Mixed Integer Nonlinear Programming problem(Mixed Integer Non-Linear Programming,MINLP),and is often a non-convex NP-hard problem(Non-deterministic polynomial hard,NP-hard).On the other hand,a lot of research is under the premise of accurate and real-time channel information,and especially for the Internet of Vehicles(Io V).But high-speed vehicles realize the complex channel characteristic,proposing challenges to estimate the channel.We firstly analyze the complex channel characteristic in the internet of vehicles.We propose the weighted average based mean framed slot estimation for highway scene,and wavelet reconstruction enabled-weighted average based mean frame slot estimation for the urban scene based on their characteristics respectively.Then we summarize common characteristics of nodes distribution in highway and urban scenes,and propose channel combination characteristic analysis based time slot average estimation scheme for general scene.Simulation results show that proposed schemes balance the relationship between the estimated error and time overhead,and provide a accurate judgment basis for offloading and frame slot allocation strategy.On the basis of obtaining accurate environmental input,the optimization problem modeled in this thesis is to maximize the sum of the throughput of all nodes in the coverage range of the base station,which is also the non-convex problem of mixed integer programming.We propose a two-stage strategy.First,the dual training sets-reinforcement learning framework is proposed to solve the offloading strategy,which can constantly adapt to the quickly varing topological characteristics and complex channel characteristics.The introduction of the dual-training set can improve the sampling probability of certain scenes and achieve the fast convergence.Secondly,we solve framed slot ratio and framed slot sequence.The offloading strategy transforms the mixed integer non-convex problem into a convex problem,and the framed slot ratio is solved by Lagrange multiplier method.The offloading queue is established based on the channel characteristics to determine the frame slot sequence,which further improves the throughput of the system.Simulation results show that the proposed framework can achieve near-optimal performance and have a time advantage over the coordinate descent method.