Data collection completed and first forecast models developed (03/2022)
After the requirements and process analysis carried out at the start of the project, in which relevant areas of application and design requirements of an AI-based ETA forecast were collected with the involvement of various representatives of inland navigation, the first half of the project duration was characterised by the procurement, processing and analysis of comprehensive databases. For the period from 2017 to 2019, movement data on historical vessel movements was obtained in the form of Inland AIS data for two inland navigation corridors. These are the corridors Rhine-Main area - ARA ports (via Main, Rhine) and Berlin - Hamburg (via Mittellandkanal, Elbeseitenkanal, Elbe) with the corresponding intermediate ports, which comprise the pilot relations of the project and allow an investigation of natural and artificial waterways (see figure). In addition to the AIS data, several public data sources such as weather data, river levels and notices-to-skippers (NtS) on infrastructure disruptions and ice conditions were obtained for Germany and the Netherlands. It was also possible to obtain data on lock opening times, geo-data on waterways and port terminals as well as company-specific data, especially from the port and terminal operators involved in the project.
Parallel to the data acquisition, activities were started to develop the ML models for the ETA forecast. For this purpose, the overall forecast problem was initially segmented into sub-problems analogous to the sub-processes of the multi-link process chain. Another central activity was the identification of suitable input variables (features) for the individual subproblems. Within the framework of an agile development approach, forecasting models have so far been implemented for the sub-problems of forecasting transit times and the Estimated Time of Departure (ETD) in seaports based on them, as well as forecasting the travel time between ports, including lock processes, intermediate stops for loading / unloading and travel interruptions. For the latter, several approaches for segmenting the process chain were tested. Several ML methods were also tested during the development with regard to the highest quality of results, including Artificial Neural Networks, Gradient Boosting, Random Forest and Support Vector Machines. In addition to the development of a model for the lead time in inland ports, the existing models are currently being improved incrementally, e.g. by taking into account additional data or features.
In addition to the development of the forecast, work was carried out in parallel on the decision assistant based on it. In the first step, ETA-relevant disruption scenarios in inland navigation, e.g. the delay of a ship at the seaport, were identified together with the practice partners, as well as suitable action alternatives for the actors involved. This information is currently being implemented on the IT side in the form of a knowledge-based system. In order to integrate the individual ETA functionalities into an overall system, a demonstrator with a practicable user interface is also currently being developed.
Further information on the SELECT project (procedure, partial results, etc.) can be found in this interim report on forecast development and in the following publications:
- Poschmann, P.; Weinke, M.; Straube, F. (2022): „Predicting Estimated Arrival Times in Logistics using Machine Learning“. In: Wang, J. (Hrsg.): Encyclopedia of Data Science and Machine Learning. Hershey, USA: IGI Global, S. 2683-2701
- Poschmann, P.; Weinke., M.; Straube, F.; Kliewer, J.; Gerhardt, F. (2022): „Künstliche Intelligenz in der Binnenschifffahrt: Steigerung der Zuverlässigkeit von Binnenschifftransporten durch datenbasierte Ankunftszeitprognosen". In: Internationales Verkehrswesen (74) 2/2022. [Veröffentlichung im Mai 2022]
- Poschmann, P.; Weinke, M.; Straube, F. (2021): „Predictions of Disruptions in Multi-Modal Transport Chains Using Artificial Intelligence". In: Bundesvereinigung Logistik (BVL) e.V. (Hrsg.): International Scientific Symposium on Logistics - Conference Volume. Bremen, S. 85-90.
- Teßmann, G. (2021): „Machine Learning: KI-basierte Entscheidungsassistenz für Logistikketten der Binnenschifffahrt". In: Schiffahrt, Hafen, Bahn und Technik (06 / 2021), S. 100-101
- Pressemitteilungen zum Projekt u. a. hier:
Presentations on the SELECT project are or were given at the following events:
- Netzwerkveranstaltung Digitale Testfelder Wasserstraße (DTW) des Bundesministeriums für Digitales und Verkehr (BMDV), Berlin, 28.06.2022 [ausstehend]
- Hamburg International Conference of Logistics (HICL) der Technischen Universität Hamburg (TUHH), 23.09.2021
- International Scientific Symposium on Logistics (ISSL) der Bundesvereinigung Logistik (BVL), 15.06.2021
Project launched in March 2020 (03/2020)
Under the leadership of the Chair of Logistics at TU Berlin, the SELECT project ("Smart decision support for inland navigation logistics chains through ETA forecasts") started on 1.3.2020. It is funded for 3 years as part of the "Innovative Port Technologies" (IHATEC) initiative of the Federal Ministry of Transport and Digital Infrastructure (BMVI): https://www.innovativehafentechnologien.de/projekt-select-gestartet/.
As part of a collaboration with various stakeholders in the German port industry (including inland shipping companies and operators of inland and seaport terminals), the SELECT project will leverage significant potential for inland navigation logistics chains from the growing data stock, including Inland AIS. With the help of artificial intelligence methods (machine learning), a digital decision assistant will be developed in SELECT that will enable inland navigation stakeholders to optimise transport routes by providing time of arrival (ETA) forecasts and related recommendations for action.
By increasing reliability, efficiency and environmental sustainability, the SELECT project contributes to increasing the competitiveness and attractiveness of inland navigation compared to alternative transport modes.
The kick-off for the project took place on 31.3.2020.