Prädiktives Leercontainermanagement für NileDutch

Einrichten einer sicheren Datenübertragung

Transmetrics bietet eine unternehmensweite VPN-Integration, die ein hohes Maß an Sicherheit für alle täglichen automatisierten Datentransfers garantiert. Nachdem Transmetrics den Integrationsprozess abgeschlossen hatte, konnte NileDutch seine Daten problemlos hochladen. Als Ergebnis standen die historischen Schifffahrtsdaten für den Zeitraum von 2011 bis 2018 zur Bereinigung und Analyse zur Verfügung. Sie enthalten mehrere Millionen Datensätze, darunter Informationen zu Buchungsrouten und Containern, Containerscans, Schiffsfahrpläne sowie Schiffs- und Container-Referenzdaten

Bereinigen und Anreichern der Daten

Ein wesentlicher Schwerpunkt der Datentransformationen für NileDutch war die Ermittlung, welche Container auf welchem Schiff zwischen jeweils zwei Häfen transportiert werden. Diese Informationen waren jedoch nicht einfach aus den historischen Daten zu extrahieren, da es einige Redundanzen in verschiedenen Tabellen gab und die Daten in ihnen teilweise widersprüchlich waren. Außerdem stammte ein großer Teil der Daten aus EDIs von Agenten an verschiedenen entfernten Standorten auf der ganzen Welt, was es schwieriger machte, die Datenqualität aufrechtzuerhalten.

Nichtsdestotrotz waren die vorherigen Investitionen von NileDutch in die Digitalisierung, die Standardisierung von Geschäftsprozessen und ein gutes TMS ein entscheidender Faktor für den Erfolg der Datentransformationen und des Gesamtprojekts. Um die Daten vollständig für die Analyse vorzubereiten, entwickelte Transmetrics zusammen mit Geschäfts- und Datenexperten von NileDutch die entsprechenden Geschäftsregeln, um die Informationen aus den verschiedenen Tabellen zu priorisieren.

Nachdem die Geschäftsregeln aufgestellt waren, bereinigte und bereicherte das Team aus Business-Analysten und Data Scientists von Transmetrics die Daten, indem es Algorithmen zur Kombination der vollständigen Historie der Containerscans mit Leerfahrten und Buchungen einsetzte. Dieser mehrstufige Prozess stellte die Vollständigkeit der Daten sicher, die dann den Schiffsplänen zugeordnet werden konnten, um festzustellen, in welchen Häfen ein Container zwischen Be- und Entladung Halt macht. Während des Datentransformationsprozesses hat Transmetrics auch die folgenden Daten weiter verbessert:

Building and testing the forecasting model

Before implementing AssetMetrics software, the planning process of global repositioning of empty containers relied on the commercial forecasts from the agents around the world, which were mainly based on the bookings and personal business experience, and later were combined into one master forecast. This approach produced satisfactory results, but for the long-term it was not optimal because of its dependency on the manual inputs and human factor. For that reason, NileDutch wanted to introduce an accurate statistical forecast.

Having extensive experience in creating proprietary AI-driven demand models for logistics companies, the team of data scientists at Transmetrics developed a highly accurate forecasting model, which was trained on several years of already cleansed and enriched historical data. Furthermore, AI and machine-learning algorithms were used to continue improving accuracy with the new data coming in. As a result, the forecasting model was predicting the releases and returns of containers for NileDutch’s entire network for the next 12 weeks on a very granular level - per each day, port, ship, and type of container. It proved to be very precise even for the periods, such as the Chinese New Year, Christmas, and the like.

In addition to the automated statistical forecast, Transmetrics and NileDutch developed the functionality that allowed all the agents in various ports to collaborate and manually suggest their predictions (e.g., when acquiring a major customer, which cannot be data-mined from the past). Combining the powerful AI-driven statistical forecast with the years of experience of the NileDutch’s agents enabled Transmetrics to create the optimal forecasting model needed for the optimization module to plan the logistics of empty containers globally.

Building and testing the optimization model

Before partnering with Transmetrics, NileDutch already had in place good processes for planning empty container repositioning, but they were time-consuming since many tasks had to be executed manually with the tools such as Excel. Complex decisions for loading and unloading empty containers to ships, repositioning them to other ports, storing them in other depots or repairing containers were taken based on the extensive experience of the Global Empty Repositioning team of NileDutch.

Such a process was manageable because the company had many containers in reserve, but despite NileDutch having skillful professionals, it is impossible even for the best planning team to calculate the optimal, lowest-cost plan. In order to do that, one would need to consider a huge amount of inputs such as how many containers would be needed in the next 12 weeks in every single port of the network on few different continents, the schedule of each ship, the free capacity of each ship and how it would change over the period, the cost to transport the containers between hundreds of possible routes, the cost to store containers in each depot, how many containers you store at the moment, which of them are within the free tier and for how many you already pay for storage, the gate costs and stevedoring costs in different ports, how many containers your customers will return in each depot over the period and so on. If some of the inputs change, the planning needs to be recalculated. Such calculations are beyond the capacity of a human brain but they could be performed with AI technologies, and that is what AssetMetrics software does.

The global optimization module was co-developed in close collaboration with NileDutch to meet all the requirements of the container shipping companies. The Transmetrics’ team started from analyzing the necessary business rules, inputs and parameters. A major focus was put on an entire set of logistics costs, which must be considered to produce a meaningful optimization.

While working on the costs, Transmetrics’ teams of data scientists and business analysts identified several data topics, which had to be addressed first in order to provide the best inputs for the optimization model: