An AUV-Assisted Data Collection Scheme for UWSNs Based on Reinforcement Learning Under the Influence of Ocean Current

Abstract

This article presents an approach to plan an autonomous underwater vehicle (AUV) data-gathering tour for underwater wireless sensor networks (UWSNs) under the influence of ocean current while ensuring data timeliness. To improve the feasibility and practicality of data collection, we propose a reinforcement learning (RL)-based AUV-assisted data collection algorithm to meet data collection time requirements, considering the effect of ocean currents on the movement of the AUV. First, graph attention network (GAT) algorithms are used to embed the information of ocean currents, time window, and sensor locations into the directed maneuver time–cost graph. Then, a 3-D AUV collection zone (3-D ACZ)-based proximal policy optimization (PPO) algorithm is used for the selection of cluster head sensors in order to further reduce the data collection delay. Finally, a new reward function is designed to generate AUV routes that satisfy the time window constraints. The proposed framework is validated through simulations with underwater environment, and the results demonstrate that the proposed method greatly improves the network lifetime and ensures that data can be collected from all sensors.