MODSIM2021, 24th International Congress on Modelling and Simulation
Modelling and Simulation Society of Australia & New Zealand
School of Science
Ships traverse the world’s oceans for a diverse range of reasons, including the bulk transportation of goods and resources, carriage of people, exploration and fishing. The size of the oceans and the fact that they connect a multitude of different countries provide challenges in ensuring the safety of vessels at sea and the prevention of illegal activities. To assist with the tracking of ships at sea, the International Maritime Organisation stipulates the use of the Automatic Identification System (AIS) on board ships. The AIS system periodically broadcasts details of a ship’s position, speed and heading, along with other parameters corresponding to the ship’s type, size and set destination. The availability of AIS data has led to a large effort to develop automated systems which could identify and be used to prevent undesirable incidents at sea. For example, detecting when ships are in danger of colliding, running aground, engaged in illegal activity, traveling at unsafe speeds, or otherwise attempting manoeuvres that exceed their physical capabilities. Despite this interest, there is a lack of a publicly available ‘standard’ dataset that can be used to benchmark different approaches. As such, each new approach to automated maritime activity modelling is tested using a different dataset to previous work, making the comparison of technique efficacy problematic. In this paper a new public dataset of shipping tracks is introduced, containing data for four vessel types: cargo, tanker, fishing and passenger. Each track corresponds to a leg of a vessel’s journey within an area of interest located around the west coast of Australia. The tracks in the dataset have been validated according to a set of rules, consisting of journeys at minimum 10 hours long, with no missing data. The tracks cover a three-year period (2018 to 2020) and are further categorised by month, allowing for the analysis of seasonal variations in shipping. The intention of releasing this dataset is to allow researchers developing methods for maritime behaviour analysis and classification to compare their techniques on a standard set of data. As an example of how this dataset can be used, we use it to build a model of ‘expected’ behaviour trained on data for three vessel categories: cargo, tanker, and passenger vessels, using a convolutional autoencoder architecture. We then demonstrate how this model of ship behaviour can be used to test new data that was not used to build the model to determine whether a track fits the model or is an anomaly. Specifically, we verify that the behaviour of fishing vessels, whose movement patterns are quite different to those of the other three vessel types, is classified as an anomaly when presented to the trained model.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.