首页|期刊导航|四川师范大学学报(自然科学版)|面向队列稳定的多无人机辅助移动边缘计算联合优化方法

面向队列稳定的多无人机辅助移动边缘计算联合优化方法OA

Joint Optimization for Queue-Stabilized Multi-UAV-Assisted Mobile Edge Computing

中文摘要英文摘要

为应对动态环境下多无人机辅助移动边缘计算中任务随机到达和信道时变问题,在队列稳定性约束下,通过联合优化部分卸载比例、资源分配、无人机-用户关联和轨迹规划,最小化系统长期能耗.该问题被建模为随机约束下的混合整数非线性规划问题,难以通过传统优化方法直接求解.为此,提出一种融合李雅普诺夫优化、马尔可夫近似与多智能体深度确定性策略梯度(multi-agent deep deterministic policy gradi-ent,MADDPG)的混合协同优化框架:L-M-MADDPG.首先,借助李雅普诺夫优化将长期队列稳定性约束转化为逐时隙确定性优化问题;进而,针对离散关联决策,采用马尔可夫近似方法构建分布式马尔可夫链,以低复杂度方式逼近最优关联策略;最后,通过归一化MADDPG算法协同优化连续型变量,包括卸载比例、资源分配与无人机轨迹.仿真结果表明,所提算法在系统能耗与队列稳定性方面均优于现有单智能体强化学习及传统关联优化算法,验证了其有效性与鲁棒性.

To address the challenges of stochastic task arrival and time-varying channels in multi-Unmanned Aerial Vehicle(UAV)-assisted Mobile Edge Computing(MEC)systems under dynamic environments,this paper investigates the joint optimization of partial task offloading,resource allocation,UAV-user association,and UAV trajectory planning,aiming to minimize the long-term system energy consumption while maintaining queue stability.The problem is formulated as Mixed-Integer Nonlinear Programming(MINLP)problem with stochastic constraints,which is intractable by traditional optimization methods.To tackle this complexity,a hybrid collaborative optimization framework,L-M-MADDPG,is proposed,which integrates Lyapunov optimization,Markov approxi-mation,and Multi-Agent Deep Deterministic Policy Gradient(MADDPG).Firstly,Lyapunov optimization is leveraged to transform the long-term queue stability constraint into a deterministic optimization problem per time slot.Subsequently,for the discrete UAV-user association subproblem,a Markov approximation method is employed to construct a distributed Markov chain,efficiently approxi-mating the optimal association strategy with low complexity.Finally,a normalized MADDPG algorithm is utilized to cooperatively opti-mize the continuous variables,including offloading ratio,resource allocation,and UAV trajectories.Simulation results demonstrate that the proposed algorithm outperforms existing single-agent reinforcement learning and traditional association optimization algorithms in terms of both system energy consumption and queue stability,validating its effectiveness and robustness.

张文杰;牛宇菲;王静;郑艺峰

闽南师范大学计算机学院,福建漳州 363000||闽南师范大学数据科学与智能应用福建省高校重点实验室,福建漳州 363000闽南师范大学计算机学院,福建漳州 363000||闽南师范大学数据科学与智能应用福建省高校重点实验室,福建漳州 363000闽南师范大学计算机学院,福建漳州 363000||闽南师范大学数据科学与智能应用福建省高校重点实验室,福建漳州 363000闽南师范大学计算机学院,福建漳州 363000||闽南师范大学数据科学与智能应用福建省高校重点实验室,福建漳州 363000

信息技术与安全科学

移动边缘计算无人机队列稳定性动态轨迹李雅普诺夫优化MADDPG

mobile edge computing(MEC)unmanned aerial vehicle(UAV)queue stabilitydynamic trajectoryLyapunov optimizationMADDPG

《四川师范大学学报(自然科学版)》 2026 (6)

748-763,16

国家自然科学基金(62372488和62376114)和福建省自然科学基金(2025J01366、2024J011560和2022J01913)

10.3969/j.issn.1001-8395.2026.06.003

评论