What it does
Platforms 2 and 3 at Zürich Hardbrücke station become very congested at busy times. Passengers tend to crowd towards the middle of the platform. Our solution takes real time data from the station to work out which areas of the station are congested or less busy. This information is then used to direct passengers to the less busy sections of the platform. The primary way of doing this is through a new set of electronic station information boards, located at station entrances. Each board displays all of the services to stop at the station, when the train is due and, most importantly, the area of the platform they should go to for their train. The system works by directing the passenger to less crowded areas on the platform; this helps to even out the number of people in each area of the platform as the system adjusts dynamically to send people to less occupied locations. The system decides which platform area to send people to through the following process:
- Compare how crowded each platform section is
- If one section is clearly less crowded, direct passengers towards this section
- In the event that several sections are similarly crowded, passengers are sent to the section where the train will be least busy.
- If all sections are lightly loaded
There are also secondary information boards, located on the platforms, detailing how busy each section of the platform is as well as detecting how crowded the carriages on the next train are. These boards take live information from both the train and the platform data and dynamically update to suggest to passengers an optimal location.
How we built it
Firstly, we examined the provided data and extracted the necessary information (e.g. train times, platforms, crowding data points etc.)
Heatmaps were generated by iterating in a moving-window though the 1.1GB dataset and exported minute-wise snapshots. To keep the heatmap snapshots comparable, they share the same kernel density calibration.
Serverside data collection and decision making (using platform congestion and train utilization data) with RESTful API to provide direction information for clients
Challenges we ran into
Unsanitised dataset, needed to sanitize the data to get what only the required information. Relating unrelated datasets to each other Development environment not ready before event.
Accomplishments that we're proud of
Creating a real time information board. This takes live train data, both arrival time and carriage occupation. This is then combined with the live platform data and the system displays to the passengers when the train is due and where to stand on the platform based on how busy each section of both the train and platform are. This is innovative because it combines platform crowd control along with train information.
Additionally, a key requirement for the information at the station entrance was to be quick and simple for the passengers to learn where they need to go. A conscious decision was made to minimise the thought process for an individual passenger. By displaying the services, and their colours, it means the necessary information is available at a glance whereas a full graphic of the platform would require more thought to process and could cause congestion when entering the station. We believe that this key aim has been successfully accomplished.
Finally, we are proud that not only have we come up with a solution for this particular station but that our solution is easily applicable for other stations across the network and even worldwide. By doing this, it should be more straightforward to resolve similar problems in the future. Additionally, not only have we achieved the scope we have also gone beyond by additionally including the time till each train and displaying how crowded each train carriage will be. These both help have more relevant and useful information and, in the case of timing, reduce congestion by preventing long waits on the platform.
What we learned
Processing big datasets slow down your apps and computer (or it is impossible). The quality of datasets and dataset description vary. But with the mentors on hand we could solve most of the questions we had.