Answers are grounded in this paper and cite their sources.
Unmanned Aerial Vehicle (UAV) networks face significant routing challenges due to highly dynamic topologies, frequent link variations, uneven traffic distribution, and heterogeneous Quality of Service (QoS) requirements. This paper proposes a deep reinforcement learning-based load balancing routing method for heterogeneous QoS requirements in clustered UAV networks. A hybrid clustering approach combining Differential Evolution and K-means (DE-Kmeans) and a Local Adjacent Proximity (LAP) topology construction strategy are first developed to establish a stable hierarchical network structure. On this basis, a Multi-Agent Dueling Double Deep Q Network (MAD3QN) routing model performs distributed routing decisions based on local network state perception, and a Transformer-based multi-task learning scheme learns heterogeneous QoS characteristics, whose representations are integrated into the routing state space and decision process through a two-stage alternating optimization strategy, resulting in the MAD3QN-MLBR routing method. Simulation results on a clustered UAV network of 105 nodes show that, compared with the DRL-based DQN routing baseline, the proposed method reduces the maximum link utilization from approximately 0.50 to 0.47 and also lowers the average link utilization, while maintaining a path delay close to the hop-count-optimal Dijkstra benchmark and considerably lower than the utilization-oriented Dijkstra-QU benchmark, with an average routing decision time of only about 2.0 ms. These results demonstrate the effectiveness of the proposed method for clustered UAV networks at the hundred-node scale operating in highly dynamic environments.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: