Research On Task Offloading Strategy In Mobile Edge Computing

With the continuous improvement of mobile communication rates and the rapid development of the Internet of Things(Io T),the number of smart mobile devices(such as vehicle intelligent terminals,wearable lo T devices,etc.)presents an exponential growth trend.To meet the growing business needs of users,Mobile Edge Computing(MEC)can provide services such as computing and storage on the edge of the network,effectively alleviating the requirements of mobile networks for transmission delay and bandwidth.However,most lo T devices have limited computing resources and battery capacity.When processing compute-intensive and delay-sensitive applications,problems such as the slow computing speed of the Central Processing Unit(CPU)and rapid battery power loss will occur,which greatly reduces the service experience of users.To solve the problem of business interruption caused by the insufficient power of Io T devices,this thesis introduces Energy Harvesting(EH)technology to improve the endurance of devices.A single-user MEC system in a two-layer heterogeneous network scenario is studied.Time-varying communication model,task queue model,task calculation model,and energy harvesting model are established.At the same time,to relieve the computing pressure of Io T devices,this thesis proposes a Q-learning based Task Offloading Algorithm.To allow the offloading strategy to dynamically adjust the target server for task offloading and the amount of offloaded tasks based on time-varying information,the decision optimization problem is modeled as a Markov Decision Process(MDP).Then the state,action,and reward function of the system is analyzed.The simulation results show that the proposed algorithm can minimize the time delay and energy cost of task execution to optimize the computing performance of the user side.Considering the different needs of different tasks for the degree of urgency in the Io T scenario,a business model based on task execution priority is proposed.In this thesis,the application scenario is extended from single-user to multi-user.Then a multi-user heterogeneous MEC system with EH function is established.In addition,in order to achieve the on-demand energy allocation of Io T devices and alleviate the resource shortage problem of the MEC servers,a Deep Q-learning Network-based Task Offloading and Resource Allocation Algorithm is proposed.Through training and updating,the task offloading and resource allocation mechanism can adapt to the dynamic MEC environment autonomously.The simulation results show that the proposed algorithm greatly reduces the long-term average cost of task execution for all Io T devices while avoiding the curse of dimensionality.