IoT-enabled wireless sensor networks optimization based on federated reinforcement learning for enhanced performance

The rapid expansion of the Internet of Things (IoT) has significantly increased the demand for efficient data collection, with Wireless Sensor Networks (WSNs) playing a crucial role in this process. However, WSNs face inherent challenges, such as limited resources, energy constraints, and fluctuating network conditions, negatively impacting network performance and longevity. Traditional Machine Learning centralized optimization approaches struggle to cope with these issues, highlighting the need for decentralized solutions. In response, this paper introduces a novel Federated Reinforcement Learning (FRL) framework designed explicitly for IoT-enabled WSNs. The proposed framework enables distributed model training across sensor nodes, allowing them to collaboratively optimize network operations without sharing raw data, thereby preserving privacy. Key contributions of this work include dynamic model updates, robust aggregation of heterogeneous data, and an energy-efficient federated averaging algorithm tailored to WSN environments. Extensive simulations demonstrate that the proposed FRL approach significantly improves WSN performance, yielding a 13% increase in packet delivery compared to Deep Q-Network (DQN) and a 30% improvement over Reinforcement Learning-Based Routing (RLBR). Additionally, energy efficiency is enhanced by 15% and 24% compared to DQN and RLBR, respectively. These findings underscore FRL's potential to overcome traditional optimization methods' limitations and substantially enhance the efficiency and longevity of IoT-enabled WSNs.