Abnormal ship route detection based on Quick-Bundles clustering algorithm

Most of the ship abnormal trajectory detection is based on division and density clustering methods, but there are few clusters based on the whole route, the Quick-Bundles algorithm does not need to cluster the route data in segments to improve the computational efficiency, but treats the route as a whole, which can perform cluster analysis of large-scale data and the abnormality judgment of the trajectory. In the actual operation, each route is first made to have the same number of points by linear interpolation, and then the minimum mean direct flip (MDF) distance is used as the similarity measure, and different numbers of clusters are obtained by changing the threshold value to find the optimal threshold to achieve the purpose of clustering. The real AIS data set in a certain time range from Bohai Bay to Taiwan Strait is selected for validation, and the clustered center-of-mass flow lines can be obtained, and a long flow line and two short flow lines are obtained, which can be used as a preliminary determination of normal and abnormal routes, and provide a basis for the analysis of the causes of abnormal routes (short flow lines). The experimental results show that the Quick-Bundles algorithm applied to ship route detection can not only visually separate short flow lines from long flow lines, but also has better computational efficiency than previous clustering algorithms, effectively improving the accuracy and precision of abnormal route and abnormal segment confirmation.