Abstract: With the widespread application of 5G communication technology, a droneassisted cloud-edge collaborative resource scheduling method based on federated deep reinforcement learning is proposed to address three major challenges in edge computing: limited terminal computing power, stringent latency requirements, and data privacy compliance. Firstly, an end-edge-cloud collaborative architecture is constructed to jointly optimize the three-dimensional trajectory of unmanned aerial vehicle(UAV), communication resources, and offloading decisions, addressing the global optimization challenge in highly dynamic environments. Secondly, a multi-head attention (MHA) mechanism is introduced into deep reinforcement learning (DRL) to accurately capture spatiotemporal coupling features and enhance the model's representation ability and convergence speed. Finally, an asynchronous federated mechanism weighted by model freshness is designed, combined with differential privacy, to eliminate the barrel effect while ensuring data security. Simulation results show that the model convergence performance is significantly improved, system energy consumption is significantly reduced, and an optimal balance between scheduling efficiency and privacy protection is achieved.