Scholarlink

In edge computing, offloading tasks from resource-limited devices to nearby edge servers reduces latency and improves efficiency. However, within this environment, performing efficient and energy-saving task offloading remains a challenge due to limited device resources, unstable networks, and diverse demands. In existing research, traditional methods encounter scalability and adaptability issues in dynamic Internet of Things (IoT) environments. Artificial intelligence approaches have shown good adaptive capabilities. However, due to the limited computational and storage resources of edge devices, the complexity of models that can be deployed is restricted. To address these challenges, this paper proposes the DPAQN algorithm, which utilizes Deep Q-Networks (DQN) and incorporates the Progressive Rectangular Window Attention Mechanism (PRWA), aiming to achieve more efficient and energy-saving task offloading strategies. To test the effectiveness of DPAQN, we compare it with five algorithms in terms of multi-objective optimization (TEH). The results indicate that under varying numbers of users and task configurations, DPAQN’s overall TEH performance is on average 20.71%–30.39% better than the other five algorithms. Moreover, it works well across different datasets, demonstrating considerable generalization capabilities.

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

Solving the Problem of Task Offloading in Heterogeneous Edge Computing Environments Using Deep Q-Networks and Attention Mechanisms

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Huifeng Liu,
  • Jintao Li,
  • Hu Zhang,
  • Hongjing Yao,
  • Lijuan Xu,
  • Minghao Shao,
  • Hongyu Xu

摘要

In edge computing, offloading tasks from resource-limited devices to nearby edge servers reduces latency and improves efficiency. However, within this environment, performing efficient and energy-saving task offloading remains a challenge due to limited device resources, unstable networks, and diverse demands. In existing research, traditional methods encounter scalability and adaptability issues in dynamic Internet of Things (IoT) environments. Artificial intelligence approaches have shown good adaptive capabilities. However, due to the limited computational and storage resources of edge devices, the complexity of models that can be deployed is restricted. To address these challenges, this paper proposes the DPAQN algorithm, which utilizes Deep Q-Networks (DQN) and incorporates the Progressive Rectangular Window Attention Mechanism (PRWA), aiming to achieve more efficient and energy-saving task offloading strategies. To test the effectiveness of DPAQN, we compare it with five algorithms in terms of multi-objective optimization (TEH). The results indicate that under varying numbers of users and task configurations, DPAQN’s overall TEH performance is on average 20.71%–30.39% better than the other five algorithms. Moreover, it works well across different datasets, demonstrating considerable generalization capabilities.