Short-term Inland Vessel Trajectory Prediction with Encoder-Decoder Models

Accurate vessel trajectory prediction is necessary for save and efficient\nnavigation. Deep learning-based prediction models, esp. encoder-decoders, are\nrarely applied to inland navigation specifically. Approaches from the maritime\ndomain cannot directly be transferred to river navigation due to specific\ndriving behavior influencing factors. Different encoder-decoder architectures,\nincluding a transformer encoder-decoder, are compared herein for predicting the\nnext positions of inland vessels, given not only spatio-temporal information\nfrom AIS, but also river specific features. The results show that the\nreformulation of the regression task as classification problem and the\ninclusion of river specific features yield the lowest displacement errors. The\nstandard LSTM encoder-decoder outperforms the transformer encoder-decoder for\nthe data considered, but is computationally more expensive. In this study for\nthe first time a transformer-based encoder-decoder model is applied to the\nproblem of predicting the ship trajectory. Here, a feature vector using the\nriver-specific context of navigation input parameters is established. Future\nstudies can built on the proposed models, investigate the improvement of the\ncomputationally more efficient transformer, e.g. through further\nhyper-parameter optimization, and use additional river-specific information in\nthe context representation to further increase prediction accuracy.\n