Forecasting crowds to reduce train platforms occupancy during rush hour

AI Quick Summary

In major train stations, SBB faces platform crowding during rush hours, impacting safety and service quality. Crowded platforms can lead to accidents, while uncomfortable waits hinder service.Our approach aims to reduce platform crowding through a two-stage process. First, we’ll enhance forecasting accuracy using SBB’s existing data, including camera footage. We’ll extract anonymized time series of people counts from each camera to generate a hierarchical time series dataset. This dataset will be used to increase the accuracy of platform crowding forecasts.In the second stage, we’ll collaborate with SBB to build scenarios that reduce crowding. For instance, we could simulate moving train arrivals to other platforms and assess their impact.These scenarios can be used by SBB to implement incentives like in-train announcements, app notifications, or discount schemes. SBB will decide on specific incentives, but this idea creates a platform to test new measures’ effectiveness. In many major train stations, SBB faces every day the problem of very crowded platforms during rush hours. This issue has safety and quality of service implications. On the safety side, a platform crowded with inattentive users might lead to accidents. Moreover, the quality of service is severely undermined by the uncomfortable wait that passengers must endure. Our idea aims at reducing platform crowding with a two-stage approach. On one hand we aim to increase the accuracy of models for forecasting how crowded a platform is going to be. Given accurate forecasts, in a second stage, we will study scenarios to reduce crowdedness on specific platforms in partnership with SBB. The first stage requires accurate forecasting methods. SBB already collects plenty of data that could allow for accurate platform crowdedness forecasts. In the Bern train station, a system of cameras is already in place, however, images cannot be exploited directly due to privacy concerns. Our plan is to extract a time series of the number of people in each camera which would automatically generate anonymized data, addressing any privacy concern. The corpus of data provided by the whole train station generates a hierarchical dataset which can be used to generate accurate forecasts for the number of users on specific platforms by exploiting our state-of-the-art time series forecasting methods. The second stage will be conducted in close collaboration with SBB. By combining the platform occupancy forecasts with train occupancy forecasts, we can create scenarios that could reduce crowding. For example, we could simulate the effect of moving a train’s arrival to another platform and assess its effect on overall congestion. These scenarios could then be used by SBB to implement incentives to nudge user behavior such as in-train announcements, push notifications on the SBB app or more elaborate discount schemes. By analyzing users’ behavior from historical data, in conjunction with the output of the scenarios, we would also be able to provide suitable periods for conducting A/B tests of the solutions. While the specific incentives will be decided with SBB, the primary objective of this idea is to establish a platform for testing the effectiveness of new measures.

Project Idea Metadata

• Project Idea Name: Forecasting crowds to reduce train platforms occupancy during rush hour
• Date: 3/14/2025 3:32:33 PM
• Administrators:
  • Dario Azzimonti

Project Idea Description

In many major train stations, SBB faces every day the problem of very crowded platforms during rush hours. This issue has safety and quality of service implications. On the safety side, a platform crowded with inattentive users might lead to accidents. Moreover, the quality of service is severely undermined by the uncomfortable wait that passengers must endure.

Our idea aims at reducing platform crowding with a two-stage approach. On one hand we aim to increase the accuracy of models for forecasting how crowded a platform is going to be. Given accurate forecasts, in a second stage, we will study scenarios to reduce crowdedness on specific platforms in partnership with SBB.

The first stage requires accurate forecasting methods. SBB already collects plenty of data that could allow for accurate platform crowdedness forecasts. In the Bern train station, a system of cameras is already in place, however, images cannot be exploited directly due to privacy concerns. Our plan is to extract a time series of the number of people in each camera which would automatically generate anonymized data, addressing any privacy concern. The corpus of data provided by the whole train station generates a hierarchical dataset which can be used to generate accurate forecasts for the number of users on specific platforms by exploiting our state-of-the-art time series forecasting methods.

The second stage will be conducted in close collaboration with SBB. By combining the platform occupancy forecasts with train occupancy forecasts, we can create scenarios that could reduce crowding. For example, we could simulate the effect of moving a train’s arrival to another platform and assess its effect on overall congestion.

These scenarios could then be used by SBB to implement incentives to nudge user behavior such as in-train announcements, push notifications on the SBB app or more elaborate discount schemes. By analyzing users’ behavior from historical data, in conjunction with the output of the scenarios, we would also be able to provide suitable periods for conducting A/B tests of the solutions. While the specific incentives will be decided with SBB, the primary objective of this idea is to establish a platform for testing the effectiveness of new measures.

In major train stations, SBB faces platform crowding during rush hours, impacting safety and service quality. Crowded platforms can lead to accidents, while uncomfortable waits hinder service.

Our approach aims to reduce platform crowding through a two-stage process. First, we’ll enhance forecasting accuracy using SBB’s existing data, including camera footage. We’ll extract anonymized time series of people counts from each camera to generate a hierarchical time series dataset. This dataset will be used to increase the accuracy of platform crowding forecasts.

In the second stage, we’ll collaborate with SBB to build scenarios that reduce crowding. For instance, we could simulate moving train arrivals to other platforms and assess their impact.

These scenarios can be used by SBB to implement incentives like in-train announcements, app notifications, or discount schemes. SBB will decide on specific incentives, but this idea creates a platform to test new measures’ effectiveness.