부산대학교 도서관

Benefiting from the high mobility and the line-of-sight communications, unmanned aerial vehicles (UAVs) and high-altitude platform (HAP) can be, respectively, designated as the edge and cloud servers to aggregate the local and edge models in hierarchical federated learning (HFL). To enable energy-efficient HFL, we manoeuvre the trajectories and control the transmit powers of UAVs over multi-cell wireless networks. Meanwhile, as the channels are reused in different cells, inter-cell interference is inevitable during the aggregation at UAVs, leading to performance degradation of HFL. To tackle these issues, an algorithm based on multi-agent twin delayed deep deterministic policy gradient (MATD3) is proposed to minimize the overall energy consumption of UAVs during the training process. The simulation results show that the proposed MATD3-based algorithm performs much better than the baseline schemes.