Task offloading optimization in multi-UAV-enabled MEC networks with ground interference

Multi-unmanned aerial vehicle (UAV)-enabled mobile edge computing (MEC) has emerged as a promising solution for providing low-latency services in the 6G-enabled Internet of Things (IoT). However, joint trajectory planning and resource allocation in multi-UAV networks face significant challenges due to dynamic ground interference sources (GIS), high-dimensional action spaces, and stringent energy constraints of both UAVs and ground terminals (GTs). In this paper, we propose a novel collaborative optimization framework named interference-aware K-means and multi-agent soft actor-critic (IAKMSAC) to maximize the system’s task completion rate while minimizing energy consumption. Specifically, we first model a multi-UAV MEC system incorporating non-orthogonal multiple access (NOMA) and explicit GIS constraints. To address non-convexity and coupling of this problem, a two-stage hierarchical algorithm is designed. In the first stage, an interference-aware K-means algorithm is employed for periodic user association, which effectively reduces inter-cluster interference and compresses the state–action space. In the second stage, we develop a multi-agent soft actor-critic framework with a shared experience replay buffer to optimize UAV trajectories and offloading strategies in continuous action spaces. Furthermore, a composite reward function is designed to balance system energy and task completion rate. Simulation results demonstrate that the proposed IAKMSAC algorithm achieves faster convergence and a higher task completion rate, particularly in energy-constrained scenarios, validating its effectiveness in complex, real-world IoT environments.