A method for clustering ship driving styles in head-on situations using collision avoidance behaviour characteristics

Currently, there are limited researches on objective demonstration and mining methods concerning the existence of ship driving styles. This paper proposes a method for clustering ship driving styles in head-on situations using collision avoidance behaviour characteristics. Firstly, the head-on situations are screened based on relative motion parameters between ships. Secondly, the improved sliding window algorithm is employed to detect the collision avoidance decision-making moment, considering the ships’ manoeuvring performance and navigation inertia. Then, collision avoidance characteristic indicators are selected, which combine the four collision avoidance requirements of ”early, large, wide, clear” proposed by the International Regulations for Preventing Collisions at Sea (COLREGs). Finally, a combination of factor analysis and the K-means + + algorithms is utilized to effectively classify and characterize ship driving styles. Empirical findings derived from Automatic Identification System (AIS) data in the Laotieshan Waterway demonstrate that ships can be categorized into four distinct driving styles: Conservative Close-Distance Avoidance (CCDA), Delayed Low-Efficacy Avoidance (DLEA), Proactive Large-Amplitude Avoidance (PLAA) and Preventive Safe-Distance Avoidance (PSDA), which account for 50%, 26%, 15%, and 9% of the total, respectively. The proposed method provides a novel research perspective and certain practical application value in comprehending the micro-behavioural traits of ships and advancing the field of Maritime Autonomous Surface Ships (MASS). • Ship driving style was clustered using collision avoidance behaviour characteristics. • The improved sliding window algorithm combined distance and heading thresholds. • The proposed method revealed the objective existence of four ship driving styles. • The case study using real AIS data showed the validity of the proposed method.