Scholarlink

<p>In Mobile Edge Computing (MEC) systems, Unmanned Aerial Vehicles (UAVs) are increasingly utilized to assist with computational tasks for users in remote areas, but limited energy and processing capabilities present challenges, particularly for handling complex tasks. This work focuses on UAV task offloading and trajectory planning under energy constraints, emphasizing task dependencies modeled as directed acyclic graphs, where offloading strategies depend on subtask completion. The Energy-Aware Task Offloading and UAV Trajectory Optimization framework based on deep Q-learning network (ETOU-DQN) is introduced, designed for UAV resource management in edge computing. The approach incorporates prioritized experience replay and noisy networks to improve exploration and decision-making, balancing task offloading with trajectory planning based on real-time environmental conditions and task dependencies. Energy-aware scheduling is combined with adaptive path planning to ensure consistent computational support and effective energy use by the mission’s completion. Simulation results show that ETOU-DQN improves finish success by over 30% and rewards by nearly 15% under moderate energy compared to the best baseline, while reducing the episodes required to reach 90% of peak reward by about 35% relative to DQN.</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

Dependency-aware task offloading and energy optimization in UAV-assisted MEC systems

  • 期刊论文

  • 发表日期：

    2025年11月19日

查看全文
  • Zike Liang,
  • Yuanzhi Ni,
  • Hongfeng Tao,
  • Heng Li

摘要

In Mobile Edge Computing (MEC) systems, Unmanned Aerial Vehicles (UAVs) are increasingly utilized to assist with computational tasks for users in remote areas, but limited energy and processing capabilities present challenges, particularly for handling complex tasks. This work focuses on UAV task offloading and trajectory planning under energy constraints, emphasizing task dependencies modeled as directed acyclic graphs, where offloading strategies depend on subtask completion. The Energy-Aware Task Offloading and UAV Trajectory Optimization framework based on deep Q-learning network (ETOU-DQN) is introduced, designed for UAV resource management in edge computing. The approach incorporates prioritized experience replay and noisy networks to improve exploration and decision-making, balancing task offloading with trajectory planning based on real-time environmental conditions and task dependencies. Energy-aware scheduling is combined with adaptive path planning to ensure consistent computational support and effective energy use by the mission’s completion. Simulation results show that ETOU-DQN improves finish success by over 30% and rewards by nearly 15% under moderate energy compared to the best baseline, while reducing the episodes required to reach 90% of peak reward by about 35% relative to DQN.