ABSTRACT In this paper, a directed acyclic graph (DAG) task‐based computing offloading strategy is proposed aiming to minimize the delay‐energy weighted cost (DEWC) of the whole system. In this scenario, multiple unmanned aerial vehicles (UAV) act as mobile servers for processing DAG tasks with different structures from multiple user equipment (UE). By considering the connection scheduling, computing frequency allocation and UAVs' trajectories, a mixed integer nonlinear programming (MINLP) problem is formulated and solved approximately. Specifically, a greedy computing frequency allocation algorithm is proposed based on task level division, and then a heuristic connection scheduling algorithm is put forward for each UE selecting UAV with minimal DEWC. Finally, the UAVs' trajectories are optimized by using the Soft Actor‐Critic (SAC) algorithm. Our strategy DGS is compared with Proximal Policy Optimization (PPO), Deep Deterministic Policy Gradient (DDPG), Particle Swarm Optimization (PSO) and baseline algorithms. The results verified the advantages in terms of convergence and DEWC.
Tang et al. (Thu,) studied this question.