AUVs Cooperative Assignment Method under Ocean Current Disturbance Based on Graph Neural Network

To address interception prediction errors and unstable cooperative assignment caused by ocean current disturbances and target maneuver uncertainties in the initial stage of multi-AUV cooperative tracking, this paper proposes a graph neural network (GNN)-based interception prediction and cooperative assignment method for multi-AUV multi-target scenarios. A current-consistent analytical interception model is developed to reduce conservative bias and improve the accuracy of interception point and rendezvous time prediction. To handle short-term target maneuvers and prior uncertainties, a physics-constrained maneuver envelope with a time-consistent risk aggregation mechanism is introduced to enhance prediction continuity and reduce assignment switching. A prior-informed heterogeneous GNN is then constructed, in which analytical predictions, risk costs, and reachability priors are encoded as edge features to enable stable cooperative assignment. Simulation results show that the proposed method outperforms comparative approaches in prediction accuracy, assignment stability, and assignment accuracy under ocean current disturbances and maneuvering uncertainties.