This paper proposes a novel federated multi-agent deep reinforcement learning (FMADRL)-based Moving Target Defense (MTD) framework to mitigate denial-of-service (DoS) attacks on low-altitude networks in unmanned aerial vehicle (UAV) swarm environments. To address the DoS threat posed by UAVs' open wireless environment, dynamic topology, and resource constraints, we design lightweight, coordinated MTD mechanisms, including leader switching, path mutation, and frequency hopping. The defense problem is formulated as a multi-agent partially observed Markov decision process (POMDP) to capture the uncertainty of the UAV swarm under attack. Each UAV is equipped with a policy agent that autonomously selects MTD actions based on partial observations and local experience. Using a policy gradient-based algorithm, the UAVs jointly optimize their policies through reward-weighted aggregation. Simulation results show that the proposed method improves attack mitigation rates by up to 34.6%, reduces mean recovery time by up to 94.6%, and reduces energy consumption and defense costs by up to 29.3% and 98.3%, respectively, compared to state-of-the-art baselines.
