This paper proposes a machine learning–based ship speed over a ground prediction model, driven by the eXtreme Gradient Boosting (XGBoost) algorithm. The data set is acquired from a world-sailing chemical tanker with five years of full-scale measurements. The model is trained using encountered metocean environments and ship operation profiles in two scenarios: through propeller revolutions per minute (RPM) or propulsion power. This model is further combined with the particle swarm optimization algorithm to integrate a sailing time control method. It optimizes constant RPM or power operation strategy to meet the requirements of a fixed estimated time of arrival.


The uncertainty over estimated time of arrival (ETA)—that is, the expected arrival date and time of a shipment to a planned destination—is one of the main problems in the decision-making stage within the entire maritime transport chain (Valcic et al., 2011; Wang et al., 2021). This uncertainty reduces the reliability of the schedule, resulting in increased delays and decreased productivity for inland transportation (Zuidwijk and Veenstra, 2015). The ship arrival delays add to the cost of vessel operating and supply chain management, and an efficient logistic plan cannot be formulated (Vernimmen et al., 2007). Thus, it is important to accurately predict the required sailing time, and the ship speed over ground (SOG) is the most essential factor in determining the ETA (Wang et al., 2020). The inevitable ship speed decrease as a result of various weather conditions always leads ship operators to frequently revise their ETA 24 hours before arrival (Fancello et al., 2011). Reliable ship speed prediction is becoming more essential in improving marine traffic control, fleet management, and cargo handling operations (Prpić-Oršić et al., 2016).

This content is only available via PDF.
You can access this article if you purchase or spend a download.