Energy-Aware Clustered Federated Learning for Underwater Sensor Networks in Naval Surveillance

Abstract

Underwater wireless sensor networks (UWSNs) are essential for naval operations, as they are used for ob- ject monitoring (surveillance), environmental monitoring, and tactical defense. However, their deployment faces serious challenges due to the limitations of underwater acoustic communication, such as high latency, low bandwidth, high packet loss rates, and severe energy constraints. Under these conditions, conventional centralized data processing approaches are impractical, necessitating a shift toward decentralized intelli- gence. This paper presents an Energy-Aware Clustered Federated Learning (CFL) framework specifically designed for UWSNs in naval systems. The proposed approach organizes sensor nodes into logical clusters, where local models are trained and aggregated at cluster heads before being transmitted to a central unit. To extend network lifetime, an energy-conscious participation scheme is employed, ensuring that only nodes with sufficient residual energy participate in model training. Moreover, a robust median-based aggregation strategy is introduced at the cluster level to mitigate the effects of noisy and lossy underwater communi- cation. Simulations conducted under realistic underwater conditions demonstrate that the proposed CFL framework achieves a model accuracy of up to 91.2%, which is comparable to centralized learning and outperforms conventional federated learning approaches. Furthermore, CFL improves network lifetime by approximately 40% and reduces communication overhead by nearly 68%, thereby enhancing overall energy efficiency compared to traditional federated learning methods. The results also show improved ro- bustness to packet loss and communication failures, highlighting the suitability of the proposed framework for autonomous underwater operations. Overall, this work illustrates the potential of federated learning to enable intelligent, resilient, and energy-efficient underwater sensor networks, opening new opportunities for future naval and maritime applications in challenging underwater environments.