A SVM based ship collision risk assessment algorithm

• Published in: Ocean engineering Vol. 202; p. 107062
• Main Authors: Zheng, Kai, Chen, Yabo, Jiang, Yi, Qiao, Shuanghu
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 15.04.2020
• Subjects: Collision avoidance; Collision risk assessment; Ship domain; SVM; Collision risk assessment; SVM; Ship domain; Collision avoidance
• ISSN: 0029-8018; 1873-5258
• DOI: 10.1016/j.oceaneng.2020.107062

Ship collision is one of the important causes of maritime accidents. Collision risk assessment is necessary for navigator or automatic navigating system to recognize the risk when facing a specific encounter situation and make the proper decision for the following collision avoidance. This paper proposes a quantitative ship collision risk assessment algorithm that is based on support vector machine (SVM) technology. The algorithm reconsiders the margin between ship domains of own ship (OS) and the target ship (TS) that is used to model the probability of collision. Moreover, the consequence of the assumed collision is also considered in the algorithm, which is modeled by the interaction force. Thus, the continuous and quantitative collision risk can be obtained by using the states of ships so that some weaknesses of the conventional risk assessment methods such as the ship domain (SD) methods and the closest point of approach (CPA) based methods can be overcome. The effectiveness of this approach is examined by simulation studies which show the advantages of the new risk assessment algorithm. •Both ship domains of own ship and target ship are considered to solve the ship collision risk assessment problem.•Support vector machine (SVM) method was applied to model the probability of the assumed collision.•The consequence of the assumed collision was also considered in the algorithm, which is modeled by the interaction force.•The algorithm provided a continuous and quantitative collision risk assessment with respect to the states of ships.•Three encounter situations were considered in the simulations to examine the effectiveness of the new assessment algorithm.