Research on Performance Boundary Test Scenario Clustering Generation Method for Ship Autonomous Collision Avoidance Algorithm

In order to get the test scenes that can give feedback on the performance boundary of the autonomous collision avoidance algorithm, this paper extends the sample scenes by Gaussian process regression method to get the test database, and uses Mean shift and DBSCAN(Density-Based Spatial Clustering of Applications with Noise) clustering algorithm to get the scene clusters corresponding to different performance modes of the autonomous collision avoidance algorithm, and finally uses the percentile method to calculate the boundary scene pairs, and the generated boundary scene pairs are used in the virtual simulation platform. The generated boundary scene pairs are proved to be abruptly changed under the algorithm performance performance experiment in the virtual simulation platform, which fully illustrates the effectiveness of the test scenes. This method can overcome the problems of manual setting of test scenes in the prior art, such as strong limitations, poor adaptability and weak targeting, and accurately find the performance boundary of the corresponding collision avoidance algorithm, which is conducive to the generation of algorithm performance envelopes and lays a foundation for the construction of future test systems.