Research On Task Schduling Strategy In Wireless Powered Edge Computing Networks

With the rapid development of technology,the scale and diversity of Internet of things(IoT)Networks are developing at a surprising speed.While the rapid development of IoT brings convenience to people's life,it also brings with it negative issues such as insufficient computing power,expensive bandwidth and environmental pollution.On the one hand,to tackle the computation problem,mobile edge computing(MEC)has been recently proposed as a viable solution.Different from the conventional cloud computing integrated with remote central clouds,the MEC provides low latency communication services by deploying computing servers at the edge of the network.On the other hand,to tackle the environmental pollution problem,radio frequency(RF)signal based wireless power transfer(WPT)has been considered as an efficient solution to charge low-power electronic devices and to reduce damage to the environment.Thus,integrating MEC and WPT into IoT networks can provide both computation and energy resource.However,deploying an IoT system that combines MEC and WPT faces several challenges such as computation mode selection,multiuser cooperative edge computing,and task scheduling.Based on this,it is of great practical significance to study the Internet of things architecture combining mobile edge computing and wireless power transfer.The major point of this thesis is focused on task scheduling for wireless powered mobile edge computing IoT network.From the perspective of edge devices scheduling strategy,we have carried out in-depth exploration and research on the goal of minimizing network average computing delay.Our major work can be summarized as follows:Firstly,This thesis studies the scheduling optimization problem aiming at minimizing the average delay of each device under the IoT architecture that combines MEC and WPT.According to the characteristics of MEC and WPT,the communication model of the IoT under the framework of WPT-MEC is constructed,and the optimization problem of minimizing network delay is formed under the constraints of energy consumption,calculation frequency,etc.It is hard to get the optimal solution of this problem for the reason that there are a large number of combinations and time-varying variables in this problem.To tackle this problem,firstly,Jensen Inequality is used to calculate the optimal local computing frequency,so as to obtain the minimum loc al computing delay.Based on this,a reinforcement learning based algorithm is proposed to solve the optimization problem and convergence of this algorithm is proved.The simulation results show that the proposed reinforcement learning based algorithm is effective,better than the benchmark algorithm,and the gap with the global optimal solution is less than 0.9%.Furthermore,The algorithm proposed in this thesis can effectively cope with the effects of channel fading and calculation frequency change.Secondly,under the IoT architecture that combines MEC and WPT,This thesis studies the scheduling optimization problem based on the prediction of devices data quantity,aiming at minimizing the average delay of each device in the network.The global optimal solution calculated in the previous part of work is obtained on the basis of limited energy transfer time of local device.Therefore,although the result obtained by using reinforcement learning method is very close to this optimal solution,it still has optimization space.Futhermore,Based on analyzing the characteristics of collected data from the piratical network,this thesis builds an echo state network(ESN)model to predict the data volume of edge device.After that,we presents the IoT communicaton model based on data prediction under the framework of WPT-MEC,and give the calculation method to determine the energy transfer time of local-computing device based on the predicted data.Based on the IoT model,this thesis constructs the optimazation model that aiming to minimizing the average delay of network.This problem also has a large number of combinations and time-varying variables.To solve this problem,a reinforcement learning based algorithm is proposed.The simulation results show that the method to use ESN to predict the amount of devices' data is effective.Meanwhile,the scheduling algorithm based on prediction can further reduce the network average delay compared to the original algorithm and achieve better results.