Data-driven analysis for safe ship operation in ports using quantile regression based on generalized additive models and deep neural network

Abstract

Marine accidents in ports can cause loss of human life and property and have negative material and environmental impacts. In South Korea, due to a pier collision accident of a large container ship in Busan New Port of South Korea, the need for safe ship operation guidelines in ports emerged. Therefore, to support quantitative safe ship operation guidelines, ship trajectory data based on automatic information system information have been used. However, because this trajectory information is variable and uncertain due to various situations arising during a ship’s navigation, there is a limit to deriving results through traditional regression analysis. Considering the characteristics of these data, we analyzed ship trajectories through quantile regression using two models based on generalized additive models and neural networks corresponding to deep learning. Among the automatic information system information, the speed over ground, course over ground, and ship’s position were analyzed, and the model was evaluated based on quantile loss. Based on this study, it is possible to suggest safe operation guidelines for the position, speed, and course of the ship. In addition, the results of this work can be further developed as a manual for the in-port-autonomous operation of ships in the future.