Scholarlink

<p>In cloud-edge collaborative environments, optimizing vehicle task offloading for data-intensive and latency-sensitive computing tasks has garnered significant attention. As traditional cloud computing architectures can’t meet real-time demands due to vehicles’ limited resources, prompting computation offloading to Mobile Edge Computing (MEC) servers as a viable solution to enhance processing speed and service quality. However, existing approaches have limitations. Conventional Markov Decision Process (MDP) models ignore vehicle mobility-induced server-switching costs and MEC resource heterogeneity constraints, thus lacking dynamic adaptability. Strategy instability and policy oscillation occur in high-density scenarios due to state-distribution shifts from single experience pools. Additionally, static service deployment results in poor vehicle-server connectivity. To tackle these problems, the Vehicle Task Offloading based on Time Division and Dynamic Decision Making (VTO-TD3) framework is presented. It models vehicle task offloading as a Constrained Markov Decision Process (CMDP). Dual experience pools are used to enhance policy robustness by separating successful and failed offloading experiences. Additionally, a CMDP reward function considering mobility-aware resource constraints is designed for joint server-selection and resource-allocation optimization. Experiments with real-world datasets show that VTO-TD3 significantly reduces service access delay, by up to 13.17%, and also shows strong robustness in dynamic scenarios. This study offers theoretical and practical guidance for low-latency, high-reliability vehicle services in intelligent transportation systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A mobility-aware task offloading approach for vehicles in cloud-edge collaborative environments

  • 期刊论文

  • 发表日期：

    2025年9月29日

查看全文
  • Feiyan Guo,
  • Xiaoqing Luo,
  • Xingshu Liu,
  • Yangkun Luo

摘要

In cloud-edge collaborative environments, optimizing vehicle task offloading for data-intensive and latency-sensitive computing tasks has garnered significant attention. As traditional cloud computing architectures can’t meet real-time demands due to vehicles’ limited resources, prompting computation offloading to Mobile Edge Computing (MEC) servers as a viable solution to enhance processing speed and service quality. However, existing approaches have limitations. Conventional Markov Decision Process (MDP) models ignore vehicle mobility-induced server-switching costs and MEC resource heterogeneity constraints, thus lacking dynamic adaptability. Strategy instability and policy oscillation occur in high-density scenarios due to state-distribution shifts from single experience pools. Additionally, static service deployment results in poor vehicle-server connectivity. To tackle these problems, the Vehicle Task Offloading based on Time Division and Dynamic Decision Making (VTO-TD3) framework is presented. It models vehicle task offloading as a Constrained Markov Decision Process (CMDP). Dual experience pools are used to enhance policy robustness by separating successful and failed offloading experiences. Additionally, a CMDP reward function considering mobility-aware resource constraints is designed for joint server-selection and resource-allocation optimization. Experiments with real-world datasets show that VTO-TD3 significantly reduces service access delay, by up to 13.17%, and also shows strong robustness in dynamic scenarios. This study offers theoretical and practical guidance for low-latency, high-reliability vehicle services in intelligent transportation systems.