首页|基于深度强化学习的任务卸载和资源分配优化

基于深度强化学习的任务卸载和资源分配优化

扫码查看
移动边缘计算(MEC)可以在网络边缘为用户提供就近的存储和计算服务,从而为移动用户带来低能耗、低时延的优势。该文针对基于超密集网络(UDN)的多用户多MEC场景,从用户侧出发,以最小化用户计算总开销为目的,解决用户在卸载过程中的卸载决策和上传传输功率优化以及MEC计算资源分配问题。具体而言,考虑到该问题是一个具有NP-hard性质的MINLP问题,该文将该问题分解为两个子问题并通过两个阶段的方式进行求解。首先在第一个阶段设计了一种基于深度强化学习(DQN)的任务卸载决策来解决任务卸载子问题,然后在第二个阶段分别使用KKT条件以及黄金分割算法解决MEC计算资源分配和上行传输功率的优化问题。仿真结果表明,所提方案在保证用户时延约束的前提下,有效降低了用户的计算开销,提升了系统性能。
Joint Optimization of Task Offloading and Resource Allocation Based on Deep Reinforcement Learning
Mobile edge computing(MEC)can provide users with nearby storage and computing services at the edge of the network,so as to bring the advantages of low energy consumption and low delay to mobile users.Aiming at the multi-user and multi MEC scenario based on ultra-dense network(UDN),starting from the user side and aiming at minimizing the total user computing overhead,we solve the problems of user unloading decision,upload transmission power optimization and MEC computing resource allocation in the unloading process.Specifically,considering that the problem is a NP hard MINLP,we decompose the problem into two subproblems and solves it in two stages.Firstly,in the first stage,a task offloading decision based on deep reinforcement learning(DRL)is designed to solve the task unloading sub problem,and then in the second stage,KKT condition and golden section algorithm are used to solve the optimization problems of MEC computing resource allocation and uplink transmission power respectively.Simulation results show that the proposed scheme effectively reduces the user's computing overhead and improves the system performance on the premise of ensuring the user's delay constraint.

ultra-dense networkmobile edge computingtask offloadingresource allocationdeep reinforcement learning

龚亮亮、张影、张俊尧、许之琛、康彬

展开 >

国网电力科学研究院有限公司,江苏南京 210006

南京南瑞信息通信科技有限公司,江苏南京 210006

南京邮电大学物联网学院,江苏南京 210003

超密集网络 移动边缘计算 任务卸载 资源分配 深度强化学习

国家自然科学基金面上项目国家自然科学基金面上项目国家自然科学基金面上项目国家博士后面上基金江苏省高等学校自然科学研究重大项目

6217123262071255620012482020M68168420KJA510009

2024

计算机技术与发展
陕西省计算机学会

计算机技术与发展

CSTPCD
影响因子:0.621
ISSN:1673-629X
年,卷(期):2024.34(4)
  • 18