Scholarlink

<p>In edge computing, Internet of Vehicles (IoV) service providers can rent edge servers deployed near users to provide real-time computational resources. From the service providers’ perspective, Edge User Allocation (EUA) is a crucial issue. The primary goal is to maximize the allocation of edge users to edge servers while minimizing server rental costs. Therefore, efficiently allocating edge users to maximize the resource utilization of individual edge servers is essential. However, users have multiple potential server choices in adjacent overlapping service coverage areas, making it NP-hard to find an optimal allocation solution quickly. To address this challenge, we propose a novel Dual Interactive-Sequence User Allocation Model (DISUAM), which employs a dual encoder-decoder structure, optimizing the user-edge server allocation strategy through feedback from the environment. A Markov Decision Process (MDP) framework is utilized, which integrates dynamic information from users and servers, leverages attention mechanisms to capture their relationships and employs a dynamically adjusted reward mechanism to perform probabilistic matching at each time step. Extensive experiments on real-world datasets demonstrate that DISUAM effectively addresses the EUA problem, outperforming state-of-the-art baseline methods.</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

Deep reinforcement learning with attention mechanisms for edge user allocation in internet of vehicles

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Pengtao Yuan,
  • Shengchao Su,
  • Wenlong Fu

摘要

In edge computing, Internet of Vehicles (IoV) service providers can rent edge servers deployed near users to provide real-time computational resources. From the service providers’ perspective, Edge User Allocation (EUA) is a crucial issue. The primary goal is to maximize the allocation of edge users to edge servers while minimizing server rental costs. Therefore, efficiently allocating edge users to maximize the resource utilization of individual edge servers is essential. However, users have multiple potential server choices in adjacent overlapping service coverage areas, making it NP-hard to find an optimal allocation solution quickly. To address this challenge, we propose a novel Dual Interactive-Sequence User Allocation Model (DISUAM), which employs a dual encoder-decoder structure, optimizing the user-edge server allocation strategy through feedback from the environment. A Markov Decision Process (MDP) framework is utilized, which integrates dynamic information from users and servers, leverages attention mechanisms to capture their relationships and employs a dynamically adjusted reward mechanism to perform probabilistic matching at each time step. Extensive experiments on real-world datasets demonstrate that DISUAM effectively addresses the EUA problem, outperforming state-of-the-art baseline methods.