首页|基于深度确定性策略梯度的工业任务卸载策略

基于深度确定性策略梯度的工业任务卸载策略

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
工业互联网背景下的移动边缘计算(Mobile Edge Computing,MEC)通过在靠近终端的位置部署边缘服务器,将计算任务卸载到工业网络边缘,以满足任务实时响应和终端节能的需求.由于工业场景复杂性和环境动态性,卸载决策需要在满足任务时延需求的同时尽可能降低系统成本,为此提出了一个基于深度确定性策略梯度(Deep Deterministic Policy Gradient,DDPG)的工业任务卸载策略.首先构建了一个端边协同的智能工厂MEC系统模型,以降低任务总时延和系统能耗为目标,将卸载问题转化为混合整数非线性规划问题,然后设计采用DDPG算法来得到最优卸载决策,提高服务质量,最大化节约系统成本.仿真结果表明,该策略在降低时延、系统能耗和成本方面比其他方法更优.
Industrial Task Offloading Strategy Based on Deep Deterministic Policy Gradient
Mobile edge computing(MEC)in the Industrial Internet domain deploys edge servers near the terminals.It supports offloading the computation task into the industrial network edge to meet the requirements of real-time task response and terminal energy saving.Due to the complexity and dynamic nature of industrial scenarios,offloading decisions must satisfy the latency requirements of industrial applications while minimizing system costs.To tackle this challenge,we introduce an industrial task offloading strategy grounded in the deep deterministic policy gradient(DDPG)approach.This strategy aims to minimize total latency and energy consumption effectively.Firstly,an intelligent factory MEC system model for device-edge collaboration was constructed,and the offloading problem was formulated as a hybrid integer nonlinear programming;then,a DDPG algorithm was designed and proposed to optimize the objective function,further enhance the Quality of Service(QoS),and maximize the system's cost savings.Simulation results demonstrate that our proposed DDPG-based strategy outperforms other approaches in reducing latency,energy consumption,and system cost.

mobile edge computing(MEC)deep reinforcement learningtask offloadingdeep deterministic policy gradient(DDPG)

梁子豪、栗娟、刘进

展开 >

武汉工程大学计算机科学与工程学院,湖北武汉 430205

智能机器人湖北省重点实验室,湖北武汉 430205

武汉大学计算机学院,湖北武汉 430072

移动边缘计算 深度强化学习 任务卸载 深度确定性策略梯度

国家自然科学基金武汉市知识创新专项曙光项目智能机器人湖北省重点实验室(武汉工程大学)科研资助项目武汉工程大学青年教师基金武汉工程大学研究生教育创新基金

621022922023010201020440HBIRL 202204K202035CX2023301

2024

10.14188/j.1671-8836.2023.0214

武汉大学学报(理学版)
武汉大学

武汉大学学报(理学版)

CSTPCD北大核心
影响因子：0.814
ISSN：1671-8836
年,卷(期)：2024.70(3)