Case Study – Forecasting Inland Vessel ETA with Predictive Analytics

Table of Contents

About the Project

The Logistics Department at the Technical University of Berlin, headed by Prof. Dr.-Ing. Frank Straube, annually prepares 120 industrial engineering students with a focus on logistics for the demands of the business world. A high value is placed on the practical relevance of the teaching.  In the course “Supply Chain Analytics” students learn to solve real-world problems in the context of Logistics and Supply Chain Management by using data and analytical methods within extensive case studies. Usually a company is selected as a case study partner, which has some kind of internal challenges, that can be addressed by conducting an analytics initiative. The students then get six weeks to tackle the given challenges in form of a group project.

In the summer semester 2021, a case study was conducted in collaboration with the SELECT research project, where students learned to develop an arrival time prediction for inland vessels using AIS data. In the research project SELECT (“Smarte Entscheidungsassistenz für Logistikketten der Binnenschifffahrt durch ETA-Prognosen”) the Chair of Logistics at the Technische Universität Berlin together with various companies from shipping industry, including BEHALA, Deutsche Binnenreederei, Duisport, HGK and modal 3 Logistik, is developing an intelligent decision support system for inland shipping. The solution developed in the project should contribute to increasing the reliability and efficiency of inland waterway transport chains in the future. One main goal, among others, is to develop a prediction for Estimated Times of Arrival (ETA) using machine learning.

The SELECT project is funded from 2020 to 2023 as part of the Innovative Port Technologies (IHATEC) funding program of the German Federal Ministry of Transport and Digital Infrastructure (BMVI).

Data Basis

FleetMon provided the SELECT project with extensive amounts of AIS data for specific inland waterway transport corridors such as the Rhine and the Elbe. The ETA prediction should not only cover the traveling process but also locking processes and resting times along the chain as well as turnaround times in the port, which finally allows to calculate process times for complex ship journeys.

Various data sources were provided for the students to work on the task. The basis for the case study were AIS messages of approximately 150 ship voyages on the Rhine, provided by FleetMon. Other data sources included water levels, which are of particular importance for the Rhine.

Goal of the Case Study

One of the most important waterways in Europe is the Rhine. Several hundred inland vessels pass through this waterway every day. For the proper coordination of processes between shipping companies, inland ports and seaports as well as other actors, precise information on arrival times is of great importance. In the course “Supply Chain Analytics”, the students were therefore given the task of developing a predictive model for ETA prediction for ship voyages between the Rhine-Main area and the ARA ports.  This analytical capability furthermore has to be made available to relevant stakeholders in form of a decision support tool. Another task for the students was therefore to design a possible front-end in form of a dashboard mock-up to provide the ETA information.

Implementation of the Case Study

The student groups followed the CRISP-DM process (Cross Industry Standard Process for Data Mining), which is a standardized, structured procedure model for conducting Analytics Initiatives. First, the Business Problem was identified. Secondly, the students went on acquiring more relevant data, like weather or holiday data for the given areas. Subsequently, that data was thoroughly analyzed to gain more insights about the characteristics of the sample route. For this purpose, the programming language R was used, which is a free software environment for statistical computing and graphics. Following an exploratory approach, the students combed through the data in search for variables, that could be used to predict the trip duration. Are ships going slower on weekends due to amateur sailors clogging the locks? Do weather conditions have a significant impact on the course of the voyage? These and many more questions were explored in this step.

When plotting the trip duration and the vessel’s draught, for example, it was examined that depending on the vessel type (tank ship or container ship), the draught did or did not have an impact on the trip duration. Therefore, tank ships are not influenced by the level of draught at all. Cargo ships on the other hand, showed a longer trip duration when their draught level increased. However, as the data set was rather small, this result should be viewed with caution.

Another group used the detailed AIS data to identify any spots, where vessels on the given route usually take breaks. It could be observed that there are more und less popular places where the breaks can be taken. This allowed for further conclusions about the characteristics of the route to be drawn.

These and many more insights were consequently used to create forecasting algorithms with the use of several methods including linear regression and extreme gradient boosting trees. This resulted in usable prediction models, which were tested against a baseline (the mean trip duration, as well as the predicted ETA by the skippers). The results were impressive – considering the student groups only had a limited time for their analyses.

As the last step, dashboards (mock-ups) were designed as a proof-of-concept of how the developed solutions could be used by stakeholders of the shipping and port industry. The dashboards were designed to fit the user’s needs by providing them with decision support for relevant cargo handling processes.

After the six weeks passed, all student groups presented their results in front of members of the SELECT project. The case study was a great benefit for the students, but also for the research project. The students on the one hand were able to analyze real-world data looking at a real business problem. They had the opportunity to deepen their skills in data analysis and predictive modeling and to gain practical insights into the work of a data analyst. The members of the SELECT project, on the other hand, received some new ideas on how the existing IT prototype being developed in the project could be further improved, for example by adding further data sources. During the supervision of the case study, it became clear, that AIS Data is an essential source of data for analyzing and optimizing logistics processes in supply chains involving inland vessel transport.


Table of Contents

Share This Project

Share on facebook
Share on twitter
Share on linkedin
Share on whatsapp
Share on email