Multi-scale collision risk estimation for maritime traffic in complex port waters.

Ship collision risk estimation is an essential component of intelligent maritime surveillance systems. Traditional risk estimation approaches, which can only analyze traffic risk in one specific scale, reveal a significant challenge in quantifying the collision risk of a traffic scenario from different spatial scales. This is detrimental to understanding the traffic situations and supporting effective anti-collision decision-making, particularly as maritime traffic complexity grows and autonomous ships emerge. In this study, a systematic multi-scale collision risk estimation approach is newly developed to capture traffic conflict patterns under different spatial scales. It extends the application of the complex network theory and a node deletion method to quantify the interactions and dependencies among multiple ships within encounter scenarios, enabling collision risk to be evaluated at any spatial scale. Meanwhile, an advanced graph-based clustering framework is introduced to search for the optimal spatial scales for risk evaluation. Extensive numerical experiments based on AIS data in Ningbo_Zhoushan Port are implemented to evaluate the model performance. Experimental results reveal that the proposed approach can strengthen maritime situational awareness, identify high-risk areas and support strategic maritime safety management. This work therefore sheds light on improving the intelligent levels of maritime surveillance and promoting maritime traffic automation.