Research On Task Offloading Algorithm Of Mobile Edge Computing For Simultaneous Wireless Information And Power Transfer

With the growth of Io T user requirements and the development of related applications,the computing requirements,bandwidth requirements,and storage requirements of mobile devices have also developed.However,limited by battery capacity and computing power,the contradiction between the ever-increasing computing needs of users and the limited resources of the equipment has become a significant problem.Therefore,mobile Edge Computing and Simultaneous Wireless Information and Power Transfer came into being to solve the problems mentioned above,and they have broad research significance and application prospects.Most researchers focus on users and design task offloading strategies based on users' experience in the existing research.However,ISPs(Internet Service Providers)as network operators also need to consider benefits.This dissertation focuses on the benefits of operators and establishes a corresponding physical model for the task offloading strategy in SWIPT to maximize the energy utilization rate of operators.In terms of the solution method,to solve the problem of high complexity in traditional algorithms,the self-learning feature of the machine learning network is introduced,and the deep reinforcement learning algorithm based on the Actor-Critic framework is used to solve the problem,thereby increasing the speed of the solution and improving the understanding quality.Finally,compared with the partial uninstall algorithm and the full uninstall algorithm,the processing rate of the proposed uninstall algorithm is as less as 45.8% and 64.2%,respectively.As the number of users increases,the limitations of the single-cell scenario become prominent.Therefore,this dissertation proposes a task offloading model of SWIPT in a multi-cell scenario.Aiming at the coupling of the optimization problem,a single base station is regarded as a single Agent,and each Agent uses the DDPG algorithm to deal with its continuous action space problem to solve its corresponding MDP problem.In addition,the model is reconstructed into a collaborative Agent Markov game model,and multi-Agent reinforcement learning is designed as its solution.Through simulation comparison and analysis,the performance of the proposed multi-Agent algorithm is at least a 12% improvement than that of the single-Agent algorithm.