无人机辅助的双层深度强化学习任务卸载算法.

UAV has the characteristics of high mobility and easy deployment, which can improve the performance of edge computing system. In order to solve the problems of UAV trajectory optimization, user power allocation and task offloading strategy, this paper proposed a Two-layer Deep Reinforcement Learning (TDRL) algorithm for task offloading. The upper layer used the multi-agent deep reinforcement learning to optimize the trajectories of UAVs, and dynamically allocated the user transmission power to improve the transmission rate of the network. The lower layer used multiple parallel deep neural networks to generate the optimal offloading decision to minimize network latency and energy consumption. The simulation results show that the proposed algorithm enables UAVs to track user movement, significantly reduces system latency and energy consumption, and provides users with better task offloading services. [ABSTRACT FROM AUTHOR]