Research Article
Research on Resource Management Algorithms for Three-Layer Collaboration in UAV-Assisted MEC
@INPROCEEDINGS{10.1007/978-3-032-33662-0_27, author={Yang Lou and Pu Zhao and Yu Hong and Li Wang}, title={Research on Resource Management Algorithms for Three-Layer Collaboration in UAV-Assisted MEC}, proceedings={Communications and Networking. 20th EAI International Conference, ChinaCom 2025, Xi’an, China, November 8--9, 2025, Proceedings}, proceedings_a={CHINACOM}, year={2026}, month={10}, keywords={UAV-assisted MEC Hierarchical Federated Learning Client Selection Resource Allocation}, doi={10.1007/978-3-032-33662-0_27} }- Yang Lou
Pu Zhao
Yu Hong
Li Wang
Year: 2026
Research on Resource Management Algorithms for Three-Layer Collaboration in UAV-Assisted MEC
CHINACOM
Springer
DOI: 10.1007/978-3-032-33662-0_27
Abstract
Mobile Edge Computing (MEC) enhances service capability by offloading computation to the network edge, providing high bandwidth and low latency. Unmanned Aerial Vehicles (UAVs), with flexible deployment and three-dimensional mobility, further extend MEC coverage. Meanwhile, Federated Learning (FL) enables distributed training with privacy preservation, making it a promising tool for UAV-assisted MEC. However, large-scale deployments face two issues: (i) device heterogeneity in data distribution, communication quality, and computational capacity complicates efficient client selection; and (ii) the coupling among communication and computation resources makes joint allocation difficult. To address these issues, we propose a UAV-assisted MEC framework based on Hierarchical Federated Learning (HFL). First, we design an Adaptive Scoring and Bidirectional Proposal Mechanism (AS-BPM) Based client selection Algorithm by jointly considering data volume, channel quality, and model staleness, with adaptive weight adjustment. Second, we develop a resource allocation algorithm using Deep Deterministic Policy Gradient (DDPG), which jointly optimizes UAV scheduling, transmission power, and computing frequency to minimize latency and energy consumption. Experiments on the CIFAR-10 dataset under both IID and non-IID settings demonstrate that the proposed framework achieves up to 83% model accuracy while reducing the total system cost by as much as 19.3% compared with baseline schemes.