Energy-Efficient DDPG-Based UAV-Assisted Asynchronous Federated Learning with MC-NOMA in IoT Networks

초록

In recent times, federated learning (FL) has gained significant interest and investigations in massive data Internet of Things (IoT) networks, primarily for its ability to enhance security and privacy. However, Unmanned Aerial Vehicle (UAV)-based FL-IoT systems present several challenges, significant reasons are the UAV mobility and resource management constraints. This research focuses on improving energy efficiency in multicarrier non-orthogonal multiple access (MC-NOMA)-based UAV-assisted FL-IoT systems. In particular, it addresses joint optimization of UAV movement, central processing unit (CPU) frequency allocation, and transmit power control while ensuring compliance with latency constraints. To achieve this goal, we propose an energy-efficient deep deterministic policy gradient (EEDDPG) algorithm to minimize the overall energy consumption of user devices in UAV-assisted FL-IoT networks. Extensive simulations demonstrate that the proposed approach consistently achieves stable convergence and significantly outperforms traditional methods in terms of energy efficiency.

키워드