Hamburger Hafen und Logistik AG (HHLA) has announced it is “one of the first ports worldwide to develop solutions for its Hamburg container terminals that use machine learning (ML) to predict the dwell time of a container at the terminal. The first two projects have now been successfully integrated and implemented into the IT landscape at Container Terminals Altenwerder (CTA) and Burchardkai (CTB)”.
At CTA, the productivity of the ASC blocks “will be increased by means of an ML-based forecast,” HHLA stated. “The goal is to predict the precise pickup time of a container. Processes are substantially optimised when a steel box does not need to be unnecessarily restacked during its dwell time in the yard. When a container is stored in the yard, its pickup time is frequently still unknown. In future, the computer will calculate the probable container dwell time. It uses an algorithm based on historic data which continually optimises itself using state-of-the-art machine learning methods”.
The project at CTB was a separate collaboration between INFORM and HPC Hamburg Port Consulting to implement INFORM’s Syncrotess Machine Learning (ML) Module. “INFORM’s AI solution predicts the dwell time (i.e., the time period the container is expected to be stored in the yard) and the outbound mode of transport (e.g., rail, truck, vessel) – both of which are crucial criteria for selecting an optimized container storage location within the yard that avoids unnecessary rehandles,” INFORM explains.
Using machine learning to predict dwell time and the outbound mode is interesting as HHLA does have data on both variables. The onward mode is submitted in the pre-announcement of the container arrival, and HHLA requires truck drivers to make an appointment to collect a container, and appointments can be booked up to three days in advance.
In remarks to WorldCargo News INFORM said its Syncrotess Machine Learning Module is not used for predicting when trucks arrive, but when containers will leave and validating/correcting the outbound transport mode. “The refinement of the outbound transport mode enables a more accurate dwell time assessment and in turn a better overall yard stack configuration”, INFORM stated.
With regard to truck appointment data, INFORM said the truck booking system “is connected to the system that feeds INFORM with the pre-announcement of the container, which entails the necessary data (i.e., Outbound Transport Mode, Estimated Time of Departure). If a booking for an outgoing container has been made upon arrival at the terminal, it is considered by the terminal’s Yard Optimiser.”
Booking data is not always accurate, and INFORM said machine learning can produce better outcomes. “The Syncrotess Machine Learning Module checks against the booking data. The Yard Optimiser, where the outcome of the Machine Learning Module is fed into, always makes live decisions whenever a container arrives/has to be moved. As such, the basis for that is improved. When data is missing, it fills the gap. When data seems to be wrong, it checks how good the machine learning prediction is and the base data information is overruled by the machine learning outcome data”.
As to how well the Machine Learning module works, INFORM said in a 2019 study it estimated it could achieve a relative improvement in prediction accuracy of 26% for dwell time predictions and 33% for an outbound mode of transport predictions. “INFORM has found that the accuracy of predictions at CTB is above 80% (i.e., in more than 80% of the predictions for outbound transport mode the Syncrotess Machine Learning Module predicts the actual outbound transport mode correctly),” the company stated.
Dr. Eva Savelsberg, SVP of INFORM’s Logistic Division said, “AI and machine learning allows us to leverage data from our past performance to inform us about how best to approach our future operations – our ML Module gives our Operations Research based algorithms the best footing for making complex decisions about what to do in the future. INFORM’s Machine Learning Module allows CTB to leverage insights generated from algorithms that continuously learn from historical data.”