This work presents a data-driven approach for the automated risk estimation of the voyage of a vessel or ship. While the industry is moving from a compliance-based framework with existing rules to a risk-based one, there is also a need to monitor the risk of a vessel from the perspective of the navigation. This is of even higher importance for the case of autonomous ships. Built based on the state-of-the-art mathematical representation, the navigation feature, each existing voyage is transformed into a corresponding series of points in [Formula: see text]-dimensional space. During the stage of pre-processing, given a set of historical Automatic Identification System (AIS) data, those records that belong to the same vessel within a certain period of time are taken as a voyage and mapped to the corresponding space of the navigation feature. After the pre-processing and during the online monitoring, the current trajectory of the vessel is transformed into the corresponding representation in the same way. Based on a nearest-neighbor search scheme, the distance from the nearest neighbor is taken as the risk of the current voyage. In other words, the deviation from the closest route in the historical data is taken as the risk. The developed method has demonstrated encouraging performance on a set of challenging historical AIS data from the Australian Maritime Safety Authority, covering three regions in the Australian territory, namely, the Bass Strait, the Great Australian Bight and the North West.
I-LSH: I/O Efficient c-Approximate Nearest Neighbor Search in High-Dimensional Space