Abstract: Traditional Mobile Edge Computing(MEC) faces challenges such as difficult base station deployment,limited coverage,and insufficient energy supply,making it hard to meet dynamic computational demands.To tackle these issues,an edge computing system integrating UAV assistance and energy harvesting devices was proposed for joint optimization of task offloading and UAV trajectory planning.With flexible deployment and high mobility,the Unmanned Aerial Vehicle(UAV) as mobile servers to provide dynamic support for devices,while energy harvesting devices ensured stable services by capturing renewable energy.The joint optimization problem was formulated as a Markov Decision Process(MDP),and a hybrid optimization algorithm framework based on deep reinforcement learning was proposed,which used Deep Q-Network(DQN) for discrete actions and Deep Deterministic Policy Gradient(DDPG) for continuous actions).Simulation results showed that the proposed algorithm achieved a better balance between energy consumption and latency,significantly reducing system costs compared to baseline methods.

Key words: unmanned aerial vehicle assisted computing, energy harvesting, task offloading, mobile edge computing

摘要： 传统移动边缘计算因基站部署困难、覆盖有限及能量不足,难以满足动态环境下的计算需求。为此,提出一种结合无人机辅助与能量收集设备的边缘计算系统,进行任务卸载与无人机路径规划的联合优化。其中,无人机凭借灵活部署和高机动性,作为移动服务器为设备提供动态计算支持;能量收集设备通过捕获可再生能源,解决传统边缘节点能量不足的问题,实现稳定高效的计算服务。将联合优化问题形式化为马尔可夫决策过程,并提出一种基于深度强化学习的混合优化算法框架。该框架基于深度Q网络(DQN)算法与深度确定性策略梯度(DDPG)算法,利用DQN处理离散动作,DDPG处理连续动作。仿真实验表明,所提算法在能耗与时延之间实现了更优的平衡,与基线方法相比显著降低了系统成本。

关键词: 无人机辅助计算, 能量收集, 任务卸载, 移动边缘计算