AI: Better scheduling for trains and trucks

Locomotive of Deutsche Bahn

When will an individual container reach the container ship at the seaport? To date, this could only be very generally predicted. With the help of artificial intelligence (AI), a model can now detect rail and road disruptions at an early stage and forecast the arrival time with a high level of accuracy; for the first time, strategic risk management along the transport route is now possible. The system was developed in the research project SMECS (Smart Event Forecast for Seaports) with DB Cargo, led by the TU Berlin and with the participation of Kühne Logistics University (KLU).

For the forecast model, the science teams examined the connections from Leipzig, Regensburg and Munich to the seaport in Hamburg. With the help of expert interviews, they first created a process map that describes in detail all physical processes in the export process from the shipper to the ship. Historical data for four years, covering 50,000 train journeys, 96,000 truck journeys and 8.6 million container-related events, was then collected and adjusted.

    Historical data train algorithm

    "We have learned from the truth of the past," says KLU Project Manager Dr. André Ludwig, Associate Professor of Computer Science in Logistics, with regard to the method. "A learning algorithm was fed and trained with historical data such as departure times and many other influencing factors. Internal influences such as the rail network’s capacity utilization, and available personnel, are just as important as external factors like switch faults, weather or police operations. This enabled us to predict when the train would arrive and then compare our forecast with the reality." The science teams identified roughly 50 relevant influencing factors, which helped them achieve 86 percent accuracy. "This means that we have clearly achieved our goal in this applied research project," Prof. Ludwig reports.

      “The ability to generate live forecasts for freight trains’ planned arrival times plays a key role for DB Cargo,” stresses DB Cargo project manager Dr. Hannah Richta. “This enables us to use resources more efficiently and increase the quality of service for our customers. Thanks to the excellent collaboration in the research project with the TU Berlin and KLU, we have gained a better understanding of the most important influencing factors and identified possible starting points for the proactive control of arrival times.”

      AI-based decision support

      Intermodal transport networks, which combine several means of transport such as train, truck and ship, are dynamic, complex and therefore vulnerable to disruptions. According to DB Cargo, every fourth freight train is delayed, by 23 hours on average (Freight Transport Forum 2015). And trucks stuck in traffic jams also generate costs in the supply chain. Yet long reaction times and inefficient ad hoc solutions run counter to the increasing demands on delivery services and delivery times.

      By integrating all the players involved, delays and terminal congestion can now be better predicted. In the event of a disruption, the AI-based system suggests alternatives to optimally control the flow of goods. As Prof. Ludwig summarizes: "Supply chains will become more cost-effective, more reliable and more competitive. This shows the huge potential of transparent data management and machine learning."

      The project was funded by Innovative Port Technologies (Innovative Hafentechnologien, IHATEC), an initiative of the Federal Ministry of Transport and Digital Infrastructure (BMVI). The project team consisted of Prof. André Ludwig and Prof. Hanno Friedrich in leading positions, together with Senior Researcher Dr. Andreas Balster.

      More information: