摘要：

针对异构网络中非均匀分布式流量的平衡问题，构建一个基于非正交多址接入-终端直连的多无人机（UAV）辅助边缘计算系统。该系统中的能耗直接受同信道干扰、计算资源和传输功率的影响，通过联合优化卸载决策、任务量、资源分配以及UAV的飞行轨迹以最小化系统加权能耗。由于所提优化问题为非凸问题且高度耦合，因此提出一种基于李雅普诺夫（Lyapunov）的两阶段在线资源协调分配方案进行求解。首先，运用Lyapunov优化理论对系统模型进行改进，以消除其对未知信息的依赖，将目标优化问题转化为仅依赖于当前时隙的优化问题；其次，将优化问题分解为4个子问题，采用交替迭代的方法进行求解，在子问题的求解过程中，采用启发式用户匹配算法获取用户最佳匹配方案，并引入改进的自适应下降交替方向乘子法来获取最优卸载决策；最后，通过连续凸逼近技术将无人机的飞行轨迹问题转化为可解的凸问题。仿真结果表明，与Local、Random、ADMM这3种基准方案相比，该方案在保证队列稳定性的前提下，能耗约降低40%~70%。

关键词: 移动边缘计算, 计算卸载, 轨迹优化, 资源分配, 能耗

Abstract:

To address the balance problem of non-uniform distributed traffic in heterogeneous networks, multiple Unmanned Aerial Vehicles(UAV)-assisted edge computing system based on Non-Orthogonal Multiple Access(NOMA)-Device-to-device(D2D) is constructed. The energy consumption in this system is directly affected by factors such as co-channel interference, computational resources, and transmission power. By jointly optimizing offloading decisions, task volume, resource allocation, and UAV flight trajectory, the system's weighted energy consumption is minimized. Because of the non-convex and highly coupled nature of the proposed optimization problem, a two-stage online resource coordination allocation scheme based on Lyapunov is proposed for the solution. First, the optimization model is improved using Lyapunov optimization theory to eliminate its dependence on unknown information and transform the target optimization problem into an optimization problem that only relies on the current time slot. Secondly, the optimization problem is decomposed into four sub problems and solved using an alternating iteration method. During the solving process of the sub problems, a heuristic user matching algorithm is used to obtain the best user matching scheme, and an improved Adaptive Descent Alternating Direction Multiplier Method(AD-ADMM) is introduced to obtain the optimal unloading decision. Finally, the flight trajectory problem of UAV is transformed into a solvable convex problem through Successive Convex Approximation(SCA) technology. The simulation results demonstrate that compared with the three benchmark schemes of Local, Random, and Alternating Direction Method of Multipliers(ADMM), this scheme reduces energy consumption by approximately 40%-70% while ensuring queue stability.

Key words: Mobile Edge Computing(MEC), computation offloading, trajectory optimization, resource allocation, energy consumption