| Author: | Chen, Dong |
| Title: | The destination port prediction for tramp ships based on AIS trajectory data mining : a case study of VLCC |
| Advisors: | Yang, Dong (LMS) |
| Degree: | M.Phil. |
| Year: | 2022 |
| Subject: | Tramp shipping -- Management -- Mathematics Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Logistics and Maritime Studies |
| Pages: | vii, 71 pages : color illustrations |
| Language: | English |
| Abstract: | Tramp shipping accounts for more than 75% of the total tonnages of ships in the market. Different from the liner shipping, tramp shipping has no fixed route, schedule, and destination port. These characteristics lead to the supply and demand imbalance that is recognized as the spatial-temporal heterogeneity problem in transportation. Destination port prediction is fundamental and significant to solve this problem. Even though AIS (Automatic Identification System) can provide the destination port information, about 70% of the information is wrong. Hence, the vessel trajectory-based method has risen to prominence, which is available for any stage of sailing. Port calls are important to extract the voyages and semantic information. To recognize port calls rapidly and correctly, we develop an optimized CB-SMoT algorithm with less time complexity. Compared with other algorithms, our algorithm can correctly identify 84.6% of port calls for bulk carriers and 90.63% for tanker ships. Grounded on the identified port calls of VLCCs, we construct a framework of three models as follows: Model 1 is a high order sequence of port calls model. The definition of order is the number of previous ports. We build different order sequences with port names as semantic information. Then we train the random forest (RF) classifier with the feature X of high order sequences and the label Y of destination port. The accuracy increases with the growth of order, which means richer previous ports information is beneficial for destination classification. When the order is larger than 3, the accuracy can reach 0.80 and above. Model 2 is a trajectory similarity model. We adopt the TRACLUS algorithm to produce representative trajectories. The representative trajectory is an extracted standard trajectory for a route but does not really exist. We calculate the similarity between the sailing trajectory and representative trajectories by SSPD (Symmetrized Segment-Path Distance) and convert them to probability matrices as semantic information. In our model, the similarity probability matrix, IMO number and DWT (Deadweight tonnage) can be the features X and the destination port can be the label Y. We train the tree-based models and find GBDT (Gradient Boosting Decision Tree) achieves the best performance. The accuracy increases along with days and exceeds 0.70 after 25 days. We also find the common segments of sailing trajectories has a negative effect on the prediction. Model 3 is a neural network model. We predict voyages for three frequent routes, respectively. The results show LSTM (Long Short-Term Memory) has the minimum RMSE (root mean squared error) of longitude and latitude for the predicted last few days' trajectory. When the time length for prediction is shorter, the correct rate is higher. The correct rate can reach 0.83 by predicting the last-48h trajectory. The reliable results can be provided two days in advance before arriving. Our work provides an innovative integrated framework of models for destination port prediction covering different stages of a voyage and gives the application guidelines of these models. In the future, based on our study, the routing optimization problem for tramp shipping will be studied. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/11999