Recognition and classification method in multi-ship encounter scenarios based on topological graph

Aiming at the problems such as the lack of characterization models, and the challenge of distinguishing between ship interactions for multi-ship encounter scenarios in complex navigable waters, a recognition and classification method was proposed based on topological graph for multi-ship encounter scenarios. Considering the dynamic characteristics of ships in time and space, automatic identification system data was divided by time slices to obtain distance data suitable for research. Based on inter-ship Haversine distances, the find-verify-and-fix clustering algorithm was applied to construct time series of topological graphs, and the representative topological graphs were automatically generated in multi-ship encounter scenarios. The similarity between different representative topological graphs for different encounter scenarios was calculated by SimGNN model, and the similarity measurement of multi-ship encounter scenarios was realized. A K-nearest neighbors classifier was employed for multi-ship encounter scenario classification, and the encounter processes of different topological graph number and various ship types were analyzed. Experimental analysis was conducted using real data from a specific day (24 hours) in Ningbo-Zhoushan water area. Research results indicate that the proposed recognition algorithm for multi-ship encounter scenarios accurately identifies 794 valid multi-ship encounter scenarios in the water area, with two-ship encounter scenarios being the most common, followed by three-ship encounter scenarios, and relatively few occurrences of four-or-more-ship encounter scenarios. This result aligns with the perception of vessel traffic service personnel. Most scenarios have a duration of less than 1 000 s, with the numbers of generated topological graphs remaining below 100, indicating a relatively similar trend in data distribution. In the same encounter scenario, there is little fluctuation in ship number, demonstrating the stability of the proposed recognition algorithm. After using the classification algorithm, there is an obvious similarity in ship types and representative topological graphs between scenarios of different durations in the same category. Scenarios of various categories are significantly different in the evolution process, duration, ship type, and representative topological graph.