Predicting ship speeds in the Arctic using deep learning on historical AIS data
MetadataShow full item record
This study offers a perspective of applying fully connected neural networks (NN) to predict vessels’ speeds on a part of the Northern Sea Route and discusses its challenges. A fully connected neural network model was used to predict vessels speeds in the Eastern Barents Sea region and the Southern Kara Sea region. The results demonstrate the ability of the model to predict the vessel’s speed based on its geographical location, time of the voyage, vessel purpose, size and ice class. The model performance was verified against randomly selected AIS (Automatic Identification System) data that were enhanced using information of the Northern Sea Route administration and of the Vessel Finder database. Testing of the model on three individual vessel transits demonstrated good results in terms of predicting general speed trends during the transits. Furthermore, we have identified two challenges in applying a fully connected NN to speed regime modelling: data quality and accessibility. These challenges are discussed and techniques to minimize them are presented in this paper. Being familiar with the advantages and limitations of fully connected NN in the modelling of vessels speeds is essential to leverage its predictive capabilities, with the goal of improving safety, emergency, and transport planning of Arctic voyages.