Ship-handling behavior pattern recognition using AIS sub-trajectory clustering analysis based on the T-SNE and spectral clustering algorithms

Abstract

Automatic identification system (AIS) trajectory data are collected from multiple sensors that record dynamic and static ship information. AIS sequences (and records) are affected by subjective ship-officer behavior such as collision-avoidance decision-making and good seamanship. Therefore, it is necessary to recognize ship-handling behavior patterns in AIS data when conducting ship-collision avoidance research and developing routing plans. Here, we propose a new method for recognizing a unique ship-handling behavior pattern based on multi-step sub-trajectory clustering analysis: (1) AIS trajectories are segmented to generate sub-trajectories and defined through 7-tuple coding; (2) 7-tuple data dimensionality reduction and data visualization are conducted using the t-distributed stochastic neighbor embedding (T-SNE) algorithm; and (3) sub-trajectory clustering based on the spectral clustering algorithm is used for behavior pattern recognition. The identified segments are used to define unique ship-handling behavior and are referred to here as ship-handling behavior basic (SHBB). This approach can help to further understand and clarify ship-handling behavior patterns while greatly improving machine learning efficiency with regard to research and planning for ship collision avoidance decision-making, route planning, and anomalous behavior detection.