Inference Model of Collision Risk Index based on Artificial Neural Network using Ship Near-Collision Data

Abstract

The judgement of Collision Risk Index (CRI) is important for safe navigation. The autonomous ship need to analyze and decide taking an action for collision avoidance. Recently, it has been possible of navigator to obtain navigation information in real time. By using such collected navigation information, many researchers have proposed inference models which are based on the fuzzy theory in order that they can make decision for safety of navigation. The conventional inference model, however, have several limitations: (i) establish a membership function with Distance of the Closest Point of Approach (DCPA) and Time to the Closest Point of Approach (TCPA) without any considerations for other ship dynamic parameters; and (ii) rely on values of simulation result using virtual navigation information. In order to overcome these limitations, we introduces the inference model with Artificial Neural Network (ANN) by learning input vector, i.e., own ship's speed, target ship's speed, own ship's course, target ship's course, bearing between own ship and target ship, distance between own ship and target ship, and target vector. Taking an actual near-collision situation into account, the proposed model can express various CRI, keeping the desirable TCPA and distance to take a proper action for collision avoidance. The proposed method conducts better decision-making than conventional ones.