DRL-Based Battery and Power Optimization in FL-enabled IoT Networks for eHealth
Alaa AlZailaa, Hao Ran Chi, Ayman Radwan, Rui L. Aguiar · 2024
In the advancing realm of e-health applications powered by 5G networks, the paramount goal of extending the sustainability of e-health services hinges on efficient energy management of Internet of Things (IoT) devices. This paper presents a cutting-edge Deep Reinforcement Learning (DRL)based approach for the dynamic optimization of IoT device selection and power allocation in Federated Learning (FL) tasks, with an emphasis on battery energy conservation to foster AIenabled e-health applications. Adhering to the strict demands of e-health applications in 5G networks, such as ultra-low latency and high data throughput, while prioritizing energy efficiency, this approach aims to prolong IoT device lifespans without sacrificing FL accuracy. By intelligently navigating network dynamics and device constraints through DRL, this framework optimizes energy usage, enhancing the sustainability and reliability of ehealth services within the 5G ecosystem. The comparative analysis reveals that the proposed approach excels in optimizing energy consumption, download and upload throughput, and training efficiency compared to conventional algorithms, establishing a new benchmark for future optimizations in FL applications across IoT devices. Compared to representative algorithms, our approach reduces energy consumption by >80% and training time by >34%, optimizing performance in FL-enabled e-Health applications.