A Data-Driven Method for Ship Collision Risk Detection in Heavy Traffic Waters

Haoran Lv, Junmin Mou, Liang Zhang, Mengxia Li, Pengfei Chen

Abstract: Intended to solve the problem of poor generalization of complex collision risk model, this paper proposes a novel collision risk model based on data-driven. Firstly, the main traffic pattern is clustered by Ordering Points to Identify the Clustering Structure algorithm. Afterwards, using probability statistics and mining the altering behaviors identify the abnormal behavior. And then, encounter situation and avoidance behaviors are identified by analyzing the relative motion characteristics of the two ships. Finally, by Kullback-Leibler dispersion and kernel density estimation methods, the key parameters are extracted from encounter data with avoidance behaviors. The collision risk model based on data-driven is constructed. To verify the effectiveness of the proposed method, the method is used in a heavy traffic area, the mouth of Yangtze River, China. The results show that the collision risk model based on data-driven can be more accurate.

Submission Number: 62