Ship trajectory uncertainty prediction based on a Gaussian Process model

Abstract

The uncertainty of ship trajectory prediction is addressed. In particular, a probabilistic trajectory prediction model is proposed that describes the uncertainty in future positions along the ship trajectories by continuous probability distributions. The ship motion prediction is decomposed into lateral and longitudinal directions, and position probabilities are calculated along these two directions. A data-driven non-parametric Bayesian model based on a Gaussian Process is proposed to describe the lateral motion uncertainty, while the longitudinal uncertainty results from the uncertainty on the ship acceleration along the route. The parameters of the probabilistic models are derived off-line based on historic trajectory information provided by Automatic Identification System (AIS) data. The model is then applied to predict the trajectory uncertainty in real time by iteratively updating the prior probability models based on new observed AIS data. Moreover, a sequential Cholesky decomposition algorithm is applied in this study to reduce the computational effort required by the Gaussian Process modelling. Three months of AIS data are used to train and test the proposed probabilistic trajectory prediction model. The results obtained show that the proposed method has high prediction accuracy and meets the demands of real-time applications.