Research Article
Secure Federated Learning for Multi-UAV Networks: A Framework Based on Cooperative Beamforming and Participant Selection
@INPROCEEDINGS{10.4108/eai.18-12-2025.2365263, author={Xiujuan Zhang and Zhenyu Zheng and Yu Du and Xin Fan and Jin Qian and Chuanwen Luo}, title={Secure Federated Learning for Multi-UAV Networks: A Framework Based on Cooperative Beamforming and Participant Selection}, proceedings={Proceedings of the 13th International Conference on Identification, Information and Knowledge in the Internet of Things, IIKI 2025, 18-21 December 2025, Chengdu, China}, publisher={EAI}, proceedings_a={IIKI}, year={2026}, month={6}, keywords={Federated Learning Unmanned Aerial Vehicle (UAV) CB Physical Layer Security Artificial Noise}, doi={10.4108/eai.18-12-2025.2365263} }- Xiujuan Zhang
Zhenyu Zheng
Yu Du
Xin Fan
Jin Qian
Chuanwen Luo
Year: 2026
Secure Federated Learning for Multi-UAV Networks: A Framework Based on Cooperative Beamforming and Participant Selection
IIKI
EAI
DOI: 10.4108/eai.18-12-2025.2365263
Abstract
This paper addresses the severe eavesdropping threat in UAV-assisted Federated Learning (FL) networks. We propose a cooperative secure framework that utilizes a novel dynamic participant selection mechanism to partition the UAV swarm into learning and jamming groups. The jamming group employs Cooperative Beamforming (CB) to actively suppress eavesdroppers, while the learning group is optimized to balance data quality and system cost. We formulate a joint optimization problem aiming to maximize the secrecy rate while minimizing system latency and energy consumption. To solve this non-convex problem with a high-dimensional mixed-action space, we propose a Deep Reinforcement Learning (DRL) algorithm named Graph Attention Multi-Head Actor-Soft Actor-Critic (GAMA-SAC). Extensive simulations demonstrate that the proposed scheme significantly enhances communication security and accelerates model convergence compared to baseline approaches.
