How BLS is on track for the future – thanks to machine learning

Machine learning can’t prevent BLS Cargo from experiencing delays. But it can, at least, make more reliable predictions and thus make logistics more efficient. This is based on a big data platform that rail company BLS is using to usher in a digital future.

It’s already late in the afternoon when the freight train from Antwerp finally pulls into Frenkendorf (Basel-Landschaft) – half a day later than planned. The workers immediately begin to unload the train and load the goods into the lorries. It’s busy, but not hectic. Why? Because the train control team within BLS Cargo’s scheduling department had already predicted the delay. This allowed the terminal workers and the logistics companies responsible for fine distribution to reschedule their operations accordingly.

The system generates up to 60,000 notifications per day from more than 200 operating locations along the Rhine-Alpine Corridor, the freight route that runs between Rotterdam in the Netherlands and Genoa in Italy. These messages, known as TRFMs (TrainRunningForecastMessages), form the data base for forecasting delays. The network covers 1,250 kilometres and is traversed by more than 20,000 BLS Cargo freight trains every year. Delays are common on the busy European rail network and pose a challenge for scheduling and logistics. ‘The arrival forecasts allow us to optimise resource planning,’ says Pascal Truniger. ‘This includes scheduling the train drivers’ shifts in compliance with the provisions of the Employment Act, but also coordinating logistics companies and customers.’

The system thus boosts efficiency not only for BLS Cargo itself, but also for its partner companies and customers, who can plan how their trucks are used for the fine distribution of goods much more precisely and reduce idle times. In turn, this ensures operations are well-managed both at the terminals themselves and within BLS Cargo’s scheduling department. There’s no need to manually estimate arrival times using Excel spreadsheets – a time-consuming task. This significantly reduces the workload on dispatchers, especially at peak times.

Better planning thanks to machine learning

The ETA system (ETA: estimated time to arrival) obtains information on the trains’ current locations from BLS’ central data lake. ‘On the one hand, we have a large SAP landscape, but, on the other, we also have a lot of semi-structured and unstructured data,’ says Marcel Graf, Data Science Team Leader at BLS. ‘We worked with Swisscom to set up a data lake on Azure so we can consolidate this information from the various sources and process it in real time.’ For this project, BLS sought out a company that was familiar with SAP as well as big data in the cloud and machine learning. ‘It wasn’t easy to find a company with these skills,’ recalls Marcel Graf. ‘However, it quickly became clear during our discussions that Swisscom has the necessary capabilities in all these areas.’ The data lake now serves as the basis for data-driven digitalisation projects at BLS and is therefore an investment in the future.

The big data platform also incorporates the raw data from the measuring points along the train route for BLS Cargo’s ETA system. A machine learning model uses this data to calculate the likely arrival time. First, though, it was important to ascertain whether it was possible to predict a reliable arrival time based on this information. ‘When I’m in my garden, I can see the main axis that BLS Cargo travels along,’ says Pascal Truniger. ‘During the proof of concept phase, I’d take my laptop into the garden every now and then to check whether the train actually passed through at the time that had been calculated.’

The train arrived as scheduled. And so the dispatchers at BLS Cargo now have calculated estimates of arrival times, meaning that they no longer have to rely on experience values that may or may not be accurate.

Optimising train operations through digitalisation

Meanwhile, the freight train has been unloaded and the composition for the next transport compiled. Thanks to the ETA system, the train driver also arrived at the terminal on time. He will now take the train back north, back to Antwerp.

BLS and BLS Cargo have also planned the next steps for the data lake. ‘We hope that the route reports from the freight trains will provide information on the reasons for delays,’ says Pascal Truniger. ‘This would allow us to further optimise operations and enhance quality.’

Marcel Graf at BLS is also convinced that digitalisation and data-driven measures will improve train operations: ‘Our goal is to use machine learning and sensor data from passenger transport to help us predict and plan maintenance intervals better.’

Passengers on the Lötschberg car transport service could benefit from artificial intelligence one day, too. ‘The data on car train utilisation can help us to plan deployments and the required capacity and, by extension, reduce waiting times,’ says Marcel Graf, looking down the track to the future.