Research On Dynamic Adaptive Offloading Method Of MEC For Multi-users

The proliferation of Internet of Things technology has resulted in an exponential increase in user-generated data.To effectively alleviate the burden of huge data volumes on the cloud,the concept of Mobile Edge Computing(MEC)has been proposed by researchers.In this new computing model,systems can analyse and process data on the edge side close to the mobile user and can provide high real-time performance services to users.At this stage,research on latency-sensitive tasks and how to adapt to dynamic changes in the environment still faces challenges in meeting the diverse needs of different types of users.To address the above phenomenon,this study presents a novel approach for offloading MEC tasks,which is tailored to the specific needs of individual users.To validate the effectiveness of the proposed method,several comparative simulation experiments are conducted.The results of these experiments demonstrate the superiority of the proposed approach over existing methods.The principal study work is as follows:(1)To better adapt to dynamically fading time-varying wireless channels and to solve the problem of finite power for end-users,Wireless Power Transfer(WPT)technology is integrated with mobile edge computing to build a multi-user task offloading model of WPT-MEC,where wireless end-users obtain energy at the wireless access point for communication and computing.It also proposes a dynamic self-adaptive offloading method based on WPT-MEC,i.e.the RLDO method,which runs mainly on the MEC server and can make real-time decisions on tasks requested by users.Considering the influence of different parameters on the methodology,the experimental results simulation show that the RLDO method can adapt well to the dynamic changes of the channel state,thus reducing the time delay and energy consumption during the execution of tasks.(2)To design a ’cloud-edge’ collaborative system task offloading overhead model based on multi-vehicle users and multi-MEC servers in a connected vehicle environment,where the cloud center coordinates the computational resources and memory capacity of the entire link.The total cost of performing the task is considered as the weighted sum of latency and energy consumption,and the total cost minimisation problem is transformed into a deep reinforcement learning-based realisation problem.A deep deterministic policy gradient-based dynamic computational offloading method for ’cloud-edge’ collaboration,namely the ECDDPG method,is proposed and implemented in the edge service layer.Simulation results show that the ECDDPG method has better performance in terms of reward value and convergence,and can reduce the total cost,computational latency and energy consumption.(3)To solve the problems of how to make fast offloading decisions for vehicle user requests in dynamic environments,how to make full use of the vehicle’s own resources and minimise task execution delay,a two-layer vehicle edge computing model consisting of a terminal layer and an edge layer is designed,and a vehicle edge computing partial task offloading method,i.e.the TADDPG method,is proposed,which is mainly based on a deep deterministic policy gradient algorithm and introduces a two-actor network mechanism with a prioritized empirical replay technique to achieve the goal of minimizing latency by learning and optimizing policies and adapting to changes in the environment,and the effectiveness of the proposed method is verified through simulation experiments.This thesis proposes different task offloading methods for mobile edge computing and vehicle edge computing,using different offloading strategies for different user requirements and different scenarios,combined with the idea of deep reinforcement learning,and shows through simulation results that the proposed method can be adapted to changes in the external environment by parameterising the weights and dynamically adjusting them during the training process,thus effectively reducing the latency,energy consumption and total cost of performing the task and meeting the diverse needs of users.