Press release | 28.07.2020 | sn

Artificial Intelligence Improves Punctuality Predictions of Transports

Researchers have developed an intelligent system to predict supply chain disruptions

Is it possible to develop precise forecasts for transport chains? Answers to this question have until now been very cautious. The complexity of individual transport and transshipment processes and the number of disruptions that occur create a high level of uncertainty. This means late deliveries, a less than optimal utilization of transport capacity and complex coordination processes resulting in far-reaching economic and environmental consequences for global logistic networks. The recently completed “Smart Events Forecast for Seaports” (SMCES) project has come up with a solution for this problem based on the targeted use of artificial intelligence (AI).

Using a range of machine learning techniques, researchers have developed an intelligent system to enable an early assessment of the estimated time of arrival of containers at important processing centers such as seaports by predicting the arrival times of truck and train transports. In addition to information regarding the ETA, participating companies will also be informed of inefficiencies and disruptions in the logistics chain as well as appropriate action to take. The development of the AI system and leadership of the project is the responsibility of Professor Dr. Ing. Frank Straube of the Chair of Logistics at TU Berlin. The project was conducted in close cooperation with DB Cargo and Kühne Logistics University (KLU).

Historical data used to train algorithms

“In order to be able to make forecasts, the transport chain has to be broken down into a number of sub-processes – truck transport, transshipment to trains, train transport - for which individual IT models with different algorithms are then developed,” explains Straube. These include prediction models for road and rail transport as well as solutions for transshipment and shunting procedures at logistics hubs. “The learning algorithms are fed historical data from the last four years from a total of 15 different IT systems belonging to the participating companies. For the three transport routes selected, Leipzig, Munich and Regensburg to Hamburg, this covers 50,000 rail transports, 96,000 road transports and 8.6 million container-related status reports,” says Peter Poschmann, research associate at the Chair of Logistics. In addition to information regarding the precise progress of transports, these data also include some 50 factors ascertained in the project which influence how the individual processes operate. These include operational information such as staff deployment planning, vehicle characteristics, utilization of routes and infrastructures as well as external factors such as roadworks and trackworks and weather conditions. Using these historical data, the algorithms learned the relationships between these factors and processing times and were able to apply this knowledge to new, unfamiliar cases.

AI-based decision support

After the individual prediction models had been developed, they were integrated into a holistic system that calculates a door-to-port ETA for transports. The interaction of on average 100 models per transport order enables a very high level of prediction accuracy. Rail transport stages with journey times of over ten hours can be estimated with 86 percent accuracy. For total transports from remote areas to ports, sometimes lasting several days, the predicted times and actual times of many orders only differed by between 10 and 90 minutes, even with major disruptive factors. “The ability to create live forecasts for the scheduled arrival times of goods trains is of key importance to DB Cargo,” stresses Dr. Bernd Pahnke, SMECS project leader for DB Cargo and speaker of the board of TFG Transfracht. “The research undertaken at TU Berlin and KLU has given us a better understanding of the most important factors involved and enabled us to identify possible approaches for a proactive management of transports.”

The SMECS forecast system is complemented by an AI-based decision-support system, which automatically detects connection conflicts based on ETA predictions and provides the organizations involved with recommendations for improvements. “This enables organizations to identify potential disruptions and delays in processes before they occur and take targeted action,” says Manuel Weinke, research associate at TU Berlin’s Chair of Logistics.

Interactive testing of results using a web demonstrator

In order to make the results of the project available to the public, the forecast system was recently transfered to a web-based application in the form of a demonstrator. This gives users the option to interactively test out the potential of AI on the basis of selected historical, anonymized transports. “The project demonstrates the feasibility of AI-based forecasts and the strategic importance of data for logistics,” stresses Straube.

The SMECS project was funded between 2017 and 2020 as part of the Federal Ministry of Transport and Digital Infrastructure’s “Innovative Hafentechnologien” (IHATEC) initiative. The Chair of Logistics at TU Berlin is currently also researching the possibilities of AI-based travel-time predictions for inland navigation. Working together with shipping companies and port operators, the recently launched SELECT project is developing intelligent ETA forecast systems to provide decision support for inland navigation logistics chains. This project is similarly funded for three years by the Federal Ministry of Transport and Digital Infrastructure.

The demonstrator can be accessed via the following website: https://www.smecs-eta.de

Contact: