EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

4074. Machine Learning Approaches for improved ETA Predictions and Turnaround Times of Inland Vessels at Sea Ports

Invited abstract in session WB-62: Machine Learning and Optimization in Ports II, stream OR in Port Operations.

Wednesday, 10:30-12:00
Room: S12 (building: 101)

Authors (first author is the speaker)

1. Peter Wenzel
2. Zhong Chu
The Hong Kong Polytechnic University
3. Frederik Schulte
Transport Engineering and Logistics, Delft University of Technology

Abstract

Predicting accurate Estimated Times of Arrival (ETA) for Inland Waterway Transportation (IWT) remains a challenge to coordinate port operations and reduce waiting. Captains report ETAs on inland vessels based on the remaining distance to the destination and the captain's experience. A comprehensive literature review highlights the current methods for ocean-going vessel trajectory-based ETA prediction, ocean-going feature engineering for direct prediction, and predictions of ETA times in IWT. The experiments use historical position data (AIS) reported by vessels heading to the port of Rotterdam, as well as dimensions of the vessels, weather conditions, and water levels. Several machine learning methods using Neural Networks, Gradient-Boosted Decision Trees, and Light Gradient-Boosting Machines are proposed to predict ETA at the Port of Rotterdam. Predictions are then compared with the captain’s ETA and the Actual Arrival Times (ATA). This paper shows that the predictions of ETA times are 98% more accurate than the captain’s ETAs. Compared to the ATA, our predictions are 99% accurate. These learning-based approaches can be further used in IWT research to create learning-informed optimization and simulation to reduce vessel turnaround times. In practice, this improved accuracy of arrival time prediction will enhance the reliability of IWT and enable the port to develop an accurate terminal visit schedule for the inland barges.

Keywords

Status: accepted

Back to the list of papers