Priority tasks based average utility maximization strategy for multi-UAV assisted MEC: A deep reinforcement learning approach

Aiming at the real-time computing problems in large-scale internet of things devices (IoTDs) scenarios, a framework for terahertz (THz) -based mobile edge computing (MEC) network with multi-unmanned aerial vehicles (UAV) collaboration is proposed. In this framework, a utility model based on latency and connection scheduling is first presented. Its significance lies in enabling high-priority tasks to obtain more computing resources, thereby reducing computing latency. Then, we formulate an optimization problem that jointly optimizes connection scheduling, computing resource allocation, and UAV flight trajectories under the objective of maximizing the average utility of IoTDs. To solve this Mixed Integer Nonlinear Programming Problem (MINLP), we use Deep Reinforcement Learning (DRL) based on learning rate decay and Prioritized Experience Replay (PER) to optimize the UAVs trajectories, and design a low-time complexity heuristic algorithm to solve the connection scheduling and resolve the computing resource allocation by an iterative algorithm. Subsequently, to evaluate the performance of our proposed algorithm, we compare it with Soft Actor-Critic (SAC), Proximal Policy Optimization (PPO), Deep Deterministic Policy Gradient (DDPG), and Particle Swarm Optimization (PSO). Simulation results show that our proposed algorithm significantly improves the average utility of IoTDs and reduces the latency of high-priority tasks. Besides, our proposed algorithm has better convergence than the above algorithms.