基于DRL的无人机辅助MEC任务卸载算法

DRL-based unmanned aerial vehicle assisted MEC offloading algorithm

针对无人机辅助移动边缘计算系统中任务卸载问题,结合深度确定性策略梯度算法(deep deterministic policy gradient,DDPG)、改进优先经验回放机制、退火思想,提出一种深度强化学习卸载算法PPS-DDPG。采用部分卸载策略,在时延约束下,联合优化用户调度、资源分配以及无人机飞行轨迹,以最小化终端用户总能耗为目标建立数学模型,运用深度强化学习算法寻找最优卸载决策。通过大量仿真实验,验证了该算法能够有效降低终端能耗,在性能和收敛程度上优于基准方案。

A deep reinforcement learning unloading algorithm,PPS-DDPG,was proposed for the task unloading problem in unmanned aerial vehicle-assisted mobile edge computing systems.The deep deterministic policy gradient(DDPG)algorithm,the improved priority experience replay mechanism,and annealing techniques were combined.A partial unloading strategy was adopted and the user scheduling,the resource allocation,and drone flight trajectories were jointly optimized under time-delay constraints,aiming to minimize the total energy consumption of end-users by establishing a mathematical model.The deep reinforcement learning algorithm was used to find the optimal unloading decision.Through numerous simulation experiments,the algorithm is verified to be useful on effectively reducing terminal energy consumption and it outperforms the benchmark solution in terms of the performance and the convergence level.