车辆边缘计算中基于MATD3的任务计算卸载策略  

MATD3-Based Task Offloading Strategy in Vehicular Edge Computing

作　　者：王洪滔 林兵 卢宇 陈乔鑫[4] 李婵 WANG Hongtao;LIN Bing;LU Yu;CHEN Qiaoxin;LI Chan(College of Physics and Energy,Fujian Normal University,Fuzhou 350117,China;Fujian Key Laboratory of Network Computing and Intelligent Information Processing,Fuzhou 350116,China;Concord University College,Fujian Normal University,Fuzhou 350117,China;School of Informatics,Xiamen University,Xiamen 361105,China;School of Public Administration&Law,Fujian Agriculture and Forestry,Fuzhou 350002,China)

机构地区：[1]福建师范大学物理与能源学院,福建福州350117 [2]福建省网络计算与智能信息处理重点实验室,福建福州350116 [3]福建师范大学协和学院,福建福州350117 [4]厦门大学信息学院,福建厦门361105 [5]福建农林大学公共管理与法学院,福建福州350002

出　　处：《福建师范大学学报(自然科学版)》2025年第3期27-37,共11页Journal of Fujian Normal University:Natural Science Edition

基　　金：国家自然科学基金项目(62072108);福建省社会科学基地重大项目(FJ2022MJDZ019);福建省高校物理学学科联盟教学改革项目(FJPHYS-2022-B02);福建省科技经济融合服务平台项目(2023XRH001);福厦泉国家自主创新示范区协同创新平台项目(2022FX5);福建省高校产学合作资助项目(2022H6024、2021H6026)。

摘　　要：车辆边缘计算(vehicular edge computing,VEC)作为一种新兴的范式,通过将计算密集型、延迟敏感的新型车辆应用卸载到移动边缘计算(mobile edge computing,MEC)服务器上而受到关注,然而在动态变化的VEC环境中执行计算卸载面临着一个关键挑战。车辆应用通常由具有依赖关系的任务组成,而车辆的移动性和应用的依赖结构增加了计算任务卸载过程中的不确定性,可能导致车辆任务卸载失败率的增加以及卸载过程中的能耗的增大。在多车任务协同卸载问题中,首先建立车辆动态卸载模型,考虑VEC环境下车辆的移动性和应用的依赖结构,然后设计一种任务调度优先级算法,以减少任务的执行延迟。研究目标是确定车辆任务的最优卸载策略,从而最小化系统中任务的卸载失败率和能耗,并采用了一种基于多智能体双延迟深度确定策略梯度(multi-agent twin delayed deep deterministic policy gradient,MATD3)算法来学习计算卸载问题的有效解决方案。实验结果表明,与其他强化学习算法相比,MATD3在卸载失败率、能耗方面分别降低38.76%、22.7%。Vehicular edge computing(VEC),as an emerging paradigm,has gained significant attention for offloading computation-intensive and latency-sensitive novel vehicular applications to mobile edge computing(MEC)servers.However,performing computation offloading in the dynamically changing VEC environment presents a major challenge.Vehicular applications,often composed of interdependent tasks,face increasing uncertainty in task offloading due to vehicle mobility and application dependency structures,potentially leading to higher task offloading failure rates and increased energy consumption during offloading.This paper investigates the issue of multi-vehicle collaborative task offloading.First,a dynamic vehicle offloading model is established,considering vehicle mobility and application dependencies within the VEC environment.Then,a task scheduling priority algorithm is designed to minimize execution delays.The goal of this research is to determine the optimal offloading strategy for vehicular tasks,thereby minimizing the task offloading failure rate and energy consumption in the system.To achieve this,the multi-agent twin delayed deep deterministic policy gradient(MATD3)algorithm is utilized to learn effective solutions for vehicular computing offloading problems.Experimental results indicate that,compared with other reinforcement learning algorithms,MATD3 reduces offloading failure rates and energy consumption by 38.76%and 22.7%,respectively.

关 键 词：车辆边缘计算 计算卸载 深度强化学习 多智能体强化学习 

分 类 号：TP393.1[自动化与计算机技术—计算机应用技术]