Vessel Trajectory Prediction Based on GAN with Kolmogorov-Arnold Network and Conditional Variational Autoencoder

Vessel trajectory prediction with automatic identification system (AIS) data helps eliminate potential navigational conflicts and has received increasing attention. However, the AIS data sometimes disappear. In such cases, it is necessary to predict a vessel's future state based on historical AIS data. Predicting a vessel's future position is inherently uncertain and nonlinear, resulting in a trajectory with numerous possibilities. To address the uncertainty in predicting the vessel's trajectory position, we propose a method based on generative adversarial network with Kolmogorov-Arnold network (KAN) and conditional variational autoencoder (CVAE). We use the latent representations obtained by randomly sampling the multivariate normal distribution as the encoding of the vessel's trajectory in different states, and use a CVAE to learn the distribution of the trajectories. In addition, we integrated the KAN into the generator to capture and model the complex nonlinear relationship inherent in the trajectory data. Experiment results show that the proposed algorithm outperforms several deep learning algorithms.