Developing a Ship Collision Risk Index estimation model based on Dempster-Shafer theory. (August 2021)

• Record Type: Journal Article
• Authors: Abebe, Misganaw; Noh, Yoojeong; Seo, Chanhee; Kim, Donghyun; Lee, Inwon
• Is Part Of: Applied ocean research. Volume 113(2021)
• Publication Date: 2021-08
• Subjects: Ship collision -- Collision risk index -- Ship maneuvering -- Dempster-Shafer theory -- Machine learning model; Ocean engineering -- Periodicals; 620.416205
• DOI: 10.1016/j.apor.2021.102735
• Languages: English
• ISSNs: 0141-1187

Abstract

Highlights: Machine-learning-based CRI estimation models were developed to replace the computationally expensive D-S evidential theory to estimate the ship collision risk. Through a comparison study of various machine learning models, the GBR model is extremely accurate and requires less computational time to estimate the CRI. A ship maneuvering algorithm using the D-S theory based GBR model was developed based on COLREG rules. The proposed method yields an accurate and efficient CRI estimation and safe ship maneuvering, which can enable proper early decisions to avoid ship collisions. Abstract: A reliable and accurate evaluation of the risk of a ship collision with other vessels is crucial for the avoidance of maritime accidents. The ship operators use the collision risk index (CRI) to detect the risk of a collision and take the necessary action. However, CRI can be assessed differently depending on various operating conditions or other vessels or encounter conditions, making it difficult to calculate such risk accurately and efficiently. In this study, a new method for calculating the CRI by combining machine learning with D-S theory is proposed to increase the efficiency of the computations while preserving the prediction accuracy of the CRI. Different machine learning models have been investigated and compared based on model accuracy and computational time, and the results showed that the gradient boosting regression (GBR) model efficiently estimates the collision risk and increases the calculation speed compared to the D-S theory. Further, the effectiveness of this approach was examined by collision avoidance simulation while simultaneously satisfying the COLREG rule and making early proper decisions to avoid collisions, which shows the advantages of the proposed risk assessment model in practical application.