Delay and Load Fairness Optimization With Queuing Model in Multi-AAV Assisted MEC: A Deep Reinforcement Learning Approach

Autonomous aerial vehicles (AAV) can alleviate the computational burden on edge devices through assisted computing. However, with the increase in the number of Internet of Things Devices (IoTDs), it is essential to establish a task queue on the AAV to schedule computing tasks from IoTDs. In addition, the load fairness of AAVs should be optimized to fully utilize the computing resources. Therefore, a multi-AAV-assisted mobile edge computing (MEC) network framework based on the queuing model is proposed, which aims at optimizing the average delay of all user devices and the load fairness of AAVs. Firstly, we prove that the arrangement of tasks with different computing delays on the AAV queue can affect the user’s average delay, so a short-job-first (SJF) queuing model is proposed to minimize the average delay of users. On this basis, a joint optimization problem related to the AAV’s three-dimensional trajectory and user connection scheduling is formulated. A SJF based low-complexity connection scheduling algorithm is proposed and combined in a deep reinforcement learning (DRL) to solve this NP-hard problem. To evaluate the performance of the proposed algorithm, we compare it with deep deterministic policy gradient (DDPG), particle swarm optimization (PSO), random moving (RM), and local computing (LC). Simulation results show that our algorithm effectively reduces user average delay and enhances AAV load fairness. Finally, SJF is compared with the traditional first-come-first-served (FCFS) queuing model on different algorithms. The results indicate that the average delay of SJF is significantly lower than that of FCFS.