Blog: Friday, 27 January 2023
Shipping containers are continually shuffled and restacked in a busy container yard – it’s unavoidable. Despite, every extra movement costs energy, time and effort – and these moves don’t add any value to the business or its customers, so reducing the number of reshuffles makes sense commercially and from a sustainability point of view. It turns out that it’s not difficult to reduce the number of reshuffles in a container yard, using basic stacking algorithms and a bit of predictive power. This has been shown in a master thesis researched and written by Niels van den Berg, a recent graduate of the MSc Supply Chain Management at Rotterdam School of Management, Erasmus University (RSM), who now works as a yard strategist for Rotterdam Short Sea Terminals. His thesis has won one of three Professor Jo van Nunen Awards for 2022.
Niels’ thesis explored the possibilities of using basic container characteristics to predict the dwell time – how long each container spends at the terminal. Knowing when a container will leave the terminal enables to keep the number of reshuffles to a minimum. He found that by using basic stacking algorithms and a bit of predictive power he could create significant gains in efficiency by reducing the number of reshuffles for retrieving containers. His findings were detailed in his master thesis, entitled Improving the container stacking strategy using dwell time predictions at Rotterdam Short Sea Terminals BV.
The container yard in the research belongs to Rotterdam Short Sea Terminals BV (RST). Niels’ thesis focussed on the import yard of the terminal, simply because there is often not much information about when a container leaves the terminal. “We rely on individual trucking companies to pick up the containers and often there is little to no information when the container is picked up”. Therefore, RST and Niels agreed to focus on the import yard. First, Niels benchmarked the yard’s actual performance to see if dwell time predictions and stacking algorithms from existing academic literature could be used to improve efficiency and performance in the yard.
Niels collected data from 154,553 container moves using 15 predictors. He used statistical analysis to make a random forest model (a collection of ‘decision trees’ for classifying observations) which could predict the dwell time of individual containers.
Unfortunately Niels’ master thesis deadline meant he didn’t have time to collect as much data as he wanted, like information about what each container held (which scientific research has shown to be a good predictor of the dwell time of a container). This meant that the performance of the prediction model was poor. However, he was able to use stacking algorithms from academic literature and adapt them for RST’s import yard. He used algorithms with and without predictions for dwell time, and ran multiple random data samples to test them. He found that strategies that used the (poor) dwell time predictions outperformed the strategies that did not use any dwell time information. “This was surprising, given that the performance of the prediction models was far from optimal,” said Niels. “This showed that even with poor quality predictions for dwell times, it was possible to make some significant efficiency gains.”
Niels could show that it was possible to reduce the number of reshuffles by 23.78 per cent – almost a quarter – simply by using dwell time predictions and stacking algorithms. “The most interesting finding is that you don’t need state-of-the-art computer systems to be more efficient. You can create a significant reduction in the number of reshuffles with simple statistical methods and models,” he commented. “This is particularly interesting for container terminals that don’t have the resources to invest in high-tech software.”
With enough data about container characteristics, other container terminals can use the same methodology to achieve improved performance – although the process does need a close relationship between analyst and client, advises Niels. “I think it’s really important to work closely with all the parties involved because there’s often valuable extra information stored elsewhere – perhaps with customers – that might help make better predictions for dwell times.”
If there’s enough information, then it should be possible to understand how data points affect dwell times and make data-driven decisions for locating containers so that the number of reshuffles is reduced.
Niels is about to begin implementing his methodology at RST. “From a software perspective, I need to investigate the exact possibilities for implementing the prediction model in our current Terminal Operating System.” It even might be possible to improve the predictions and say exactly how many reshuffles can be avoided in future, but co-operation with customers and other stakeholders is important for improving the predictive power of the model.
Rotterdam School of Management (RSM)
Science Communication and Media Officer
Corporate Communications & PR Manager