Predicting Vessel Tracks in Waterways Using Bi-LSTM and Transformer Models

This paper proposes an approach to predict vessel tracks in waterways using a bidirectional Long Short-Term Memory (Bi-LSTM) and a transformer model. For this purpose, the positions of the buoys along the Elbe and Weser rivers delimiting the waterway will be determined and merged with Automatic Information System (AIS) data. Additionally, tide data as well as weather information will be used to train the model. The created models are then trained with and without tide and weather data and evaluated against each other to check the influence of the additional attributes. Our results show that the lowest average error for each predicted feature is achieved with the Bi-LSTM, where no tide and weather data were used for training. Also the transformer model reaches lower prediction errors than a linear prediction, which is used as a baseline model.