Towards sustainable wireless rechargeable sensor networks: a federated multi-agent reinforcement learning approach for cooperative wireless charging

Wireless rechargeable sensor networks (WRSNs) face persistent energy limitations due to the finite battery capacity of sensor nodes, which can compromise network reliability in remote or dynamic environments. To address these challenges, this paper proposes a novel federated multi-agent reinforcement learning (FedRL-MARL) framework for adaptive and cooperative energy replenishment using multiple mobile wireless chargers (MWCs). Unlike traditional centralized approaches, FedRL-MARL leverages decentralized policy learning, enabling each MWC to train on local observations while contributing to a globally aggregated model through federated updates. The problem is formulated as a Markov decision process (MDP), allowing agents to make intelligent charging and routing decisions in real time, even in the presence of obstacles and changing node demands. Simulation results demonstrate that the proposed method improves network lifetime by up to 16%, enhances energy efficiency by over 9%, and significantly reduces communication overhead when compared to state-of-the-art approaches. This research sets a strong direction for scalable, decentralized energy replenishment in next-generation sensor networks. It lays the groundwork for resilient, efficient power management across diverse applications such as smart cities, environmental sensing and autonomous IoT deployments.