Trajectory clustering for SVR-based Time of Arrival estimation

Abstract

Accurate vessel Time of Arrival (ToA) estimation is important for port operation and resource management. In this paper, we propose a data-driven approach for estimating the ToA of ships based on Automatic Identification System (AIS) data. The proposed approach exploits a novel trajectory clustering approach to extract representative routes, from which training trajectories are grouped into clusters and ToA estimation models based on Support Vector Regressor (SVR) are trained for each cluster. The proposed ToA estimation approach adopts a short period of the latest AIS data as input and performs SVR model selection and ToA estimation in real-time. A historical AIS dataset provided by the Danish Maritime Authority is used to evaluate the proposed approach. Numerical results for tanker ships travelling to the port of Skagen demonstrate that the proposed approach can achieve ToA estimations that are 40% to 70% more accurate than state-of-the-art ToA estimation methods.