Prediction-based data collection of UAV-assisted Maritime Internet of Things

In maritime data collection scenarios, due to the constraints of wireless communication and environmental factors such as wave motion, sea surface ducting effects, and sea surface curvature, floating sensor nodes are unable to establish direct data transmission links with the base station. The advent of unmanned aerial vehicle (UAV)-assisted Maritime Internet of Things (MIoT) provides a feasible solution to this challenge. However, in existing maritime environments, floating sensor nodes drift due to ocean currents, posing significant challenges for long-distance data transmission while maintaining a low age of information (AoI). Consequently, we introduce a prediction-based UAV-assisted data collection mechanism for MIoT. In this scheme, we first select convergence nodes responsible for gathering data from floating sensor nodes and forwarding it to passing UAVs. We then propose a dynamic clustering algorithm to allocate task areas to UAVs, with each area assigned to a single UAV for data collection from floating sensor nodes. To ensure stable data offloading by UAVs, we develop a UAV relay pairing algorithm to establish reliable air-to-air relay paths and provide two data offloading modes: distal UAV and proximate UAV. Owing to the drift of floating sensor nodes influenced by ocean currents, we employ a deep echo state network to predict the positions of floating sensor nodes and utilize a multi-agent deep deterministic policy gradient to solve the UAVs trajectory planning problem. Under this mechanism, the UAVs can adaptively adjust its flight path while exploring floating sensor nodes in dynamically changing ocean sensor node scenarios. Extensive experiments demonstrate that the proposed scheme can adapt to dynamic ocean environments, achieving low-AoI data collection from floating sensor nodes.