EURO-Online login
- New to EURO? Create an account
- I forgot my username and/or my password.
- Help with cookies
(important for IE8 users)
3974. Pick-up and Drop-off Prediction for Tugboat Operations – An Inverse Optimization Approach
Invited abstract in session WA-62: Machine Learning and Optimization in Ports I, stream OR in Port Operations.
Wednesday, 8:30-10:00Room: S12 (building: 101)
Authors (first author is the speaker)
1. | Frederik Schulte
|
Transport Engineering and Logistics, Delft University of Technology | |
2. | Yuanfeng Xu
|
TU Delft | |
3. | Mahnam Saeednia
|
Tu delft | |
4. | Pedro Zattoni Scroccaro
|
TU Delft |
Abstract
Historical analysis of data related to the pick-up (i.e. the connection of the tugboat to the vessel) and drop-off (i.e. consequent disconnection) locations at ports such as the Port of Rotterdam shows a divergence from the scheduled plans, hypothetically related to factors such as weather, other scheduled activities for the tugboat, or other unknown factors. This poses challenges for the scheduling of tugboats and vessel-related activities, calling for an effective approach to predict the locations where vessels are picked up and dropped off. In this work, we propose an inverse optimization model to emulate the decisions of pilots in selecting the actual drop-off locations, aiming at learning the implicit decision models used by them. To this end, the proposed inverse optimization approach takes the actual decisions as inputs and determines an objective and/or constraints that replicate these decisions. In the model, ‘signals’ are defined as the important factors that impact the decision of pilots such as vessel-, towage-, and weather-specific information. The model is constructed based on an orienteering problem, in which the objective function aims at maximizing the total score of a path calculated as the sum of the visited node scores. The output of the model minimizes a loss function defined as the difference between the predicted and actual locations. The results enable advanced anticipatory tugboat scheduling that outperforms conventional myopic approaches.
Keywords
- Machine Learning
- Combinatorial Optimization
- Transportation
Status: accepted
Back to the list of papers