An Ordinal Model of Risk Based on Mariner's Judgement

Abstract

General rightsThis document is made available in accordance with publisher policies. Please cite only the published version using the reference above. This paper describes a statistical method for learning and estimating the risk posed by other crafts in the vicinity of a vessel and an overview to its possible spatial application, simulating how professional mariners perceive and assess such risk and using navigational data obtained from a standard integrated bridge. We propose a non-linear model for risk estimation which attempts to capture mariners' judgement. Questionnaire data has been collected that captures and quantifies Mariners' judgements of risk for crafts in the vicinity, where each craft is described by measurements that can be obtained easily from the data already present in the ship's navigational equipment. The dataset has then been used for analysis, training and validating Ordered Probit Models in order to obtain a computationally efficient data driven model for estimating the risk probability vector posed by other crafts. Finally, we discuss how this risk model can be incorporated into decision making and path finding algorithms.

Time‐Varying Risk Measurement for Ship Collision Prevention

“…the OOWs. These measurements usually assess the probability of collision between the OS and the target ships (TSs) and the measured risk usually is devoted to collision prevention for one ship Lopez-Santander & Lawry, 2016). The applications of these measurements are involved in collision alert system and path planning(Y.…”

“…In some research, collision risk is defined as subjective risk and measured based on the perception of the experts, such as the captains (Lopez-Santander & Lawry, 2016), pilots (Chin & Debnath, 2009), etc. The aim of these measurements is training a risk measurement to approach the experts' feeling of dangers in different scenarios.…”

“…The aim of these measurements is training a risk measurement to approach the experts' feeling of dangers in different scenarios. Specifically, this approach is to summarize the observed data into the values of several variables, and then it uses expert judgment based on the experience (Ahn, Rhee, & You, 2012;Lopez-Santander & Lawry, 2016) and navigation rules Lopez-Santander & Lawry, 2016) to combine these value into a single assessment of the collision risk. The collision alarm is then triggered by comparison of the risk level with a certain threshold that also relies on expert judgments (Baldauf, Benedict, Fischer, Motz, & Schröder-Hinrichs, 2011;Hilgert & Baldauf, 1997).…”

“…In a word, the measured collision risk represents the degree of experts' belief on the occurrence of the collisions. The variables that have been used including: motion related variables, such as relative position (Szlapczynski & Szlapczynska, 2016), relative speed (W. Zhang, Goerlandt, Montewka, & Kujala, 2015), relative bearing (Tam & Bucknall, 2010), types of encountering Lopez-Santander & Lawry, 2016), Distance at Closet Point of Approach (DCPA) and Time to Closet Point of Approach (TCPA) (Chin & Debnath, 2009), ship domain (Szlapczynski & Szlapczynska, 2016;Wang, 2010), etc. ; descriptions of the TSs, such as ship length (Wang, 2010), ship type , gross tonnage (Chin & Debnath, 2009), etc.…”

“…Ship collision risk measurements are usually from the perspective of one ship, denoted as Own Ship (OS), and focus on evaluating the levels of danger of approaching ships Lopez-Santander & Lawry, 2016;Tam & Bucknall, 2010). In existing measurements, the maneuver feature of the OS is neglected.…”