SNIINet: Trajectory prediction using ship navigation information interaction-aware neural network
Abstract
With the progress in artificial intelligence technology, deep learning-based ship trajectory prediction methods have been integrated into the industry, greatly enhancing prediction accuracy and aiding maritime professionals in identifying potential safety concerns. However, in congested navigation zones like ports and waterway intersections, ship navigation is influenced by unpredictable factors and nearby ships, complicating trajectory prediction. This article proposes a trajectory prediction model using ship navigation information interaction-aware neural network (SNIINet) between ships based on deep learning. This model constructs a position attention block (PAB) to encode the interrelationships between ships, taking into account the uncertain information during navigation, and uses an information fusion block (IFB) to fuse these two pieces of information. Finally, a global attention block (GAB) is used to facilitate information extraction between input and output trajectory sequences. This model is implemented in the form of an encoder-decoder structure, which utilizes long short-term memory (LSTM) networks to encode information and effectively store and propagate it through its internal state. Experiments were carried out in three different encounter scenarios across various sea areas. The results showed that the S model outperformed the baseline model in terms of prediction accuracy and generalization capability, demonstrating its effectiveness and practicality for predicting ship trajectories in complex navigation settings.