DeepCurrent: An Attention-Driven Graph Neural Network for Energy-Efficient Routing and Data Aggregation in UIoT Networks
Nellore Kapileswar, Judy Simon
Abstract
The proliferation of Underwater Internet of Things (UIoT) has enabled transformative applications in oceanography, environmental monitoring, and underwater surveillance. However, the harsh underwater environment, characterized by high latency, limited bandwidth, and rapidly depleting energy sources, imposes significant constraints on reliable communication and data transmission. This research study proposes DeepCurrent, a novel attention-driven Graph Neural Network (GNN) framework designed to optimize energy-efficient routing and data aggregation in UIoT networks. The proposed model constructs a dynamic graph representation of the UIoT topology and leverages spatial-temporal attention mechanisms to prioritize data paths based on link quality, residual energy, and hop distance. Unlike traditional routing protocols, DeepCurrent adaptively learns the optimal transmission strategy through graph message passing and multi-head attention, thus enabling context-aware, scalable, and real-time decision-making. Extensive simulations conducted using realistic underwater channel models demonstrate superior performance in terms of packet delivery ratio, network lifetime, and energy consumption when compared with state-of-the-art techniques. DeepCurrent thus provides a robust and intelligent framework for sustainable underwater communication systems.