Maritime traffic partitioning: An adaptive semi-supervised spectral regularization approach for leveraging multi-graph evolutionary traffic interactions

Maritime situational awareness (MSA) has long been a critical focus within the domain of maritime traffic surveillance and management. The increasing complexities of ship traffic, originating from sophisticated multi-attribute interactions among multiple ships, coupled with the continuous evolution of traffic dynamics, pose significant challenges in attaining accurate MSA, particularly in complex port waters. This study is dedicated to establishing an advanced methodology for partitioning maritime traffic, aimed at enhancing traffic pattern interpretability and strengthening ship anti-collision risk management. Specifically, three interaction measure metrics, including conflict criticality, spatial distance, and approaching rate, are initially introduced to quantify different aspects of spatiotemporal interactions among ships. Subsequently, a semisupervised spectral regularization framework is devised to adeptly accommodate both multiple interaction information and prior knowledge derived from historic partitioning structures. This framework facilitates the segmentation of regional traffic into multiple clusters, wherein ships with the same cluster exhibit high temporal stability, conflict connectivity, spatial compactness, and convergent motion. Meanwhile, an adaptive hyperparameter selection model is engineered to seek optimal traffic partitioning outcomes across diverse scenarios, while also incorporating user preferences for specific interaction indicators. Comprehensive experiments using AIS data from Ningbo-Zhoushan Port are undertaken to thoroughly assess the models' efficacy. Research findings from case analyses and model comparisons distinctly showcase the capability of the proposed approach to successfully deconstruct the regional traffic complexity, capture high-risk zones, and strengthen strategic maritime safety measures. Consequently, the methodology holds significant promise for advancing the intelligence of maritime surveillance systems and facilitating the automation of maritime traffic management.