Objectives To address the complex and dynamic environments faced by vessels navigating coastal waters, this paper proposes a dynamic group extraction method for ship navigation risk in coastal waters based on spectral clustering.

Methods Taking Xiamen Port waters as the study area, the following steps were implemented. First, key information such as ship positions, speeds, and navigation statuses under different scenarios at various times were extracted using automatic identification system (AIS) data, enabling real-time calculation of the spatial-temporal collision risk level between each pair of vessels. Subsequently, a vessel conflict relationship network was constructed based on the computed potential risk values, forming a topological structure to describe risk distribution and vessel interactions within the waters. Next, by integrating risk values and distances, modularity was introduced as a criterion to estimate the number of clusters. Spectral clustering was then applied to partition vessels into risk groups characterized by tightly connected intra-group conflicts and sparse inter-group relationships. Finally, a risk potential field for vessel groups was established to delineate hotspot areas, and the risk values of each group were further calculated to precisely identify these hotspots.

Results The results demonstrate that this method effectively reveals the spatial distribution characteristics of risks in complex coastal navigation environments through group clustering, enabling timely and accurate identification of risk hotspots.

Conclusions The findings will assist maritime regulatory authorities in comprehensively understanding real-time navigation dynamics and implementing preventive measures to enhance vessel traffic safety.