Crossing behavior decision-making for inland ferry ships based on Machine Learning

Abstract

The crossing behavior decision-making when encountering target ships for ferry ships is one of the key issues in enhancing traffic safety and efficiency for ferries. It is the foundation of route planning and would contribute to encountering risk assessment. The traditional collision avoidance decision-making approaches cannot be directly applied to inland ferry ships due to the characteristics in terms of high safety requirement, lower priority in collision avoidance and so on. Considering navigating characteristics and collision avoidance rules for ferries, the ferry's crossing actions can be simplified to be a binary problem that is crossing from ahead or behind target ships. So Ferries Crossing Actions Determination (FCAD) approach is proposed to quantify the relative likelihood of crossing actions. By training algorithms on ferries encountering with target ships, crossing actions can be identified. Amongst the various methods been tested, Xgboost shows good performance with Recall of 99% and Accuracy of 94%. The proposed approach contributes to improvements on ferries actions decision-making and intelligent route planning control.