Novel Maritime Traffic Analysis Techniques to Enhance Maritime Situational Awareness

Abstract

The maritime domain is characterized by complex vessel movements with intricate spatiotemporal patterns and interdependencies. However, Automatic Identification System (AIS) data, despite being a rich real-time source of vessel positioning, violates the assumption of independent and identically distributed (i.i.d.) data points due to inherent temporal, spatial, and network dependencies. Traditional data analysis techniques under such assumptions encounter limitations when applied to AIS data; ignoring dependencies in data points can result in inaccurate clustering or pattern detection, underestimation of uncertainty in predictions, and biased parameter estimates in models assuming independent observations. This thesis aims to develop advanced data-driven frameworks and methodologies that leverage time-series analysis, spatial data mining, and network science to develop a novel model for destination port prediction. The objective is to explore the potential of supporting port authorities in forecasting traffic inflow and outflow within their local environment by monitoring AIS messages. This thesis first presents a novel approach to enrich trajectory representations by integrating AIS data with port information and segmenting trajectories based on port points, thereby homogenizing vessel movement patterns. A semi-supervised clustering algorithm is then proposed for these trajectory segments, employing contextual data to derive clustering constraints. This algorithm effectively identifies preferred vessel paths, and port-to-port traffic flows directly from AIS trajectories. Building upon these clusters, a data-driven method is developed where trajectory patterns dictate the network topology. This scalable graph adapts to different geographical regions and traffic densities, eliminating the need for static route networks. Utilizing the traffic network representation and trajectory similarity measures, a prediction method is developed to forecast vessel destinations based on recent movements. Evaluations on real-world AIS datasets demonstrate promising results, with the model expressing uncertainty through probability distributions for potential destinations and dynamically updating these probabilities as the vessel progresses. This research advances maritime analytics by developing data-driven methodologies that model intricate spatiotemporal patterns and dependencies in AIS data, account for the complex connectivity of maritime traffic, and enable enhanced prediction capabilities. By overcoming the limitations of traditional techniques, this work contributes to the state-of-the-art in maritime data analytics and decision support systems.