Research On Task Offloading Decision Of Mobile Edge Computing For Multi-Users

Mobile Edge Computing(MEC)is an emerging computing paradigm that can extend the resources of cloud computing to the edge of the wireless access network close to the user.Computationally resource-constrained end devices can perform computationally intensive and latency-sensitive tasks to meet their computational and latency requirements through MEC.Computational offloading,one of the key technologies for MEC,has been well discussed in a range of existing literature.However,MEC environments are complex,dynamic,and heterogeneous,and it is challenging to solve the computational offloading and policy optimization of MEC by traditional means.In this thesis,we delve into the computational offloading decision problem in complex MEC environments,aiming to improve the overall performance of MEC systems by optimizing task offloading strategies.In this thesis,we study the strategy optimization problem of computational offloading in MEC environment,and the main works are as follows:(1)Computation offloading policy optimization under UAV-assisted mobile edge computing.Considering special scenarios such as sparse user distribution or obstructed communication,a system model of UAV-assisted MEC is constructed in this study.The system optimizes the computational offloading strategy with the optimization objective of minimizing the weighted sum of delay and energy consumption under the constraints of task offloading rate,UAV flight angle and flight speed.This joint optimization problem is formulated as a Markov decision process in the study,and a PPO-based dynamic unloading algorithm is proposed to solve this problem.A series of simulation experiments demonstrate the performance and advantages of the proposed algorithm.(2)Study of offloading decision for associative tasks in mobile edge computing environment.For the offloading problem of associative tasks in MEC scenarios,the computational offloading strategy is optimized with the optimization objective of minimizing the latency.In this study,the association rules between tasks are represented using a DAG,and the task offloading problem is modeled as an Markov Decision Process.Topological sorting is used in the study to transform the original DAG into the input of the Sequence-to-Sequence neural network to output a sequence of task offloading plans.In this study,a deep reinforcement learning algorithm is proposed to stabilize the training results in order to efficiently train this neural networks and prevent them from falling into local optima in order to optimize the computational offloading strategy.A series of simulation experiments demonstrate the performance and advantages of the proposed algorithm.