In the accident of a collision, the distance at which the other ship was first detection range was either not found or less than 1 mile, and it was investigated that the ship operator did not recognize the ship's approach until the impending collision.
As in the case of statistics, considering the situation in which the target ship is recognized from a very close distance in a collision accident, in order to prevent accidents, it is essential to secure sufficient first detection range and initial action in advance.
In this study, The Risk Informed Classification Model (RICM) was developed that can recognize the encounter situation of the target ship from a sufficient distance in order to secure a first detection range to prevent collision between ships.
For this study, a COLREG-based Encounter Classification Model (ECM) was applied to determine the ship's encounter situation, and based on the survey results in which experienced ship's officer, using the Logistic Regression(one of the machine learning methods), developed a model that provides information and warning notifications to ship operators so that they can secure a first detection range from the risk of collision with other ships.
Using Logistic Regression, developed a model 1 that judges whether collision risk information needs to be provided in advance for overtaking, crossing, head-on, and ambiguous encounter situations, and a model 2 that determines and provides information and warnings.
As a result of verifying and analyzing the RICM model, it was analyzed that all of the dependent variables were statistically significant in the length and speed of the own and target ships, DCPA, Range, and TCPA used as independent variables. In general, the larger the size of the ship, the higher the ship speed, and the closer the distance to target ships, the more subjective risk the ship operator feels tends to increase, and it was confirmed that the RICM model similarly follows.
The satisfaction of the captains who are currently receiving the RICM model collision risk information service in real time through the e-Navigation monitor was evaluated, and it was found that they were generally satisfied with the provision of collision risk information. In addition, it was suggested that it is useful in a situation of low visibility.
As a result of comparing and analyzing maritime traffic data for ships receiving collision risk information by RICM model in real time and ships using general AIS, it was analyzed that there was no distinct difference. If information and warning notifications by the RICM model were received, it was expected that the passing distance with the other ship would increase, but the difference could not be confirmed. It was confirmed as a limitation that long-term observation and data acquisition are required to evaluate the traffic effect.
This study is expected to serve as a basis for research on providing information to prevent future ship collision accidents.
Continuous RICM model effect analysis, traffic evaluation, acquisition of additional learning data, and model advancement will contribute to the prevention of ship collision accidents.