Towards Vessel Arrival Time Prediction Through a Deep Neural Network Cluster

The prediction of accurate vessel arrival times is essential and challenging at the same time to plan vessel arrivals with sufficient accuracy, coordinate berthing manoeuvres and monitor ship traffic efficiently. This paper investigates a new approach by using clusters consisting of deep artificial neural networks (DNNC). For this purpose, the considered coverage area of the Weser river was divided into geospatial domains. An also developed linear regression model (LNNC) served as a reference model, which was generated analogously to the machine learning approach on the clusters. The estimated time of arrival prediction was evaluated at a distance of 50 km between the estuary of the Weser river into the North Sea and the target industrial port. It could be shown that the mean deviation from the actual travel time at a distance of 50 km is −19.80 min for the DNNC and 67.90 min for the LNNC. At a distance of 33 km from the industrial port, the mean deviation of the DNNC decreases to 2.85 min and for the LNNC to 54.40 min. Furthermore, it has been observed that the shorter the distance to the destination port, the more accurate the predictions become.