Research On Task Offloading Strategy Of Mobile Edge Computing Based On Intelligent Learning Method

In recent years,with the rapid development of wireless communication technology,the number of mobile devices has increased sharply and been more than personal computers.While bringing convenience to people,the development of mobile devices also needs the support of technology.Because mobile devices are limited by battery capacity,their computing power is small.To mean the dependence and requirements of people,more tasks need to be completed by mobile devices.Hence,how to complete complex tasks with limited computational capability and battery capacity has become a key role.Investigators propose the concept of Mobile Edge Computing(MEC),which means the mobile devices can offload tasks to the edge computing nodes equipped with computing server by wireless communication link,and then the nodes assist the mobile devices in computing tasks and returns the results to the mobile devices.Although the wireless transmission technology has been very developed,in mobile and complex communication environment,how to efficiently offload tasks and schedule limited communication resources has attracted our attention.This thesis mainly studies the task offloading strategy and wireless resource allocation in the MEC network.There we considered a three-layer wireless transmission network with multiple relays and multiple computational access points(CAP),where users need to complete computing tasks with the assistance of CAP.To improve the communication rate,relays,which have some computational capability,are equipped between the users and CAP to help the transmission.We consider two key factor in MEC network:latency and energy consumption.Moreover,the latency and energy consumption are combined by the linear method,which not only converts the multi-objective optimization problem into single-objective optimization,but also can further adapt more scenarios.In further,we propose the task offloading strategies,bandwidth allocation schemes and bandwidth selection criteria to minimize the system cost.In order to explore other factors in the MEC network,we further studied the effect of price on mobile users.Specifically,we assume a system model with multiple users and one CAP,where the users need to complete the task with assistance of the CAP to reduce latency,and the CAP charges the user some fee for assisting the users to calculate the tasks.In the whole process,the user hope to reduce the system cost(linear combination of delay and charge)as much as possible by controlling the offloading rate of the local task,while CAP can increase its revenue by dynamically adjusting the price.Therefore,a game is formed between the user and CAP.We model the game relationship between the user and CAP through Win or Learning Fast(Wolf)in multiagent reinforcement learning,and learn to reach a stable state.Moreover,we have carried out a lot of simulation experiments,and the experimental results show the effectiveness of our proposed schemes in multi-dimension.