Here in D.C., display boards in Metro stations will tell folks when their train will arrive. Riders can check their Twitter feed or smartphone for alerts on delays as they happen. But now a mathematician in Sweden has come up with an algorithm that will be able to tell travelers about a delay up to two hours ahead of time.
According to the Daily Mail, mathematician Wilhelm Landerholm devised the algorithm by using the mountains of data collected on where trains are at any given time in a system to predict whether riders will face a delay even once a situation has been resolved. The key, it seems, is in the ripple effect. Landerholm calls his model, “The Computer Prognosis.”
The Daily Mail, explains how it all works:
The model works in a similar way to how a seismograph monitors earthquakes, looking for significant peaks.
In The Commuter Prognosis, these peaks are represented by trains pulling into a station later than the scheduled arrival time.
It automatically compares how a similar delay on the same line impacted the network on previous occasions to predict how it will affect the network in the near future.
Traffic control centres are then able to issue extra trains, or take trains out of service, to balance demand.
The Commuter Prognosis can also be used to warn commuters of delays in advance, to help them plan their journey better.
For example, if a commuter is told there is a chance there will be a delay of 10 minutes on their line in two hours’ time they can set off earlier or choose an alternative route.
By warning commuters of problems ahead of time, it might also help reduce overcrowding associated with delays or it might enable the system to send in additional trains to compensate for the delays.
Given Metro’s recent track record for on-time performance has been less than stellar, some may see value in just such an algorithm. At least one person who spotted the story on Gizmodo, however offered this snarky take on how the algorithm could be used here.
in the case of Washington, DC’s metro system we’d probably be better served with an algorithm to detect when the system will be on time since we usually add an extra 30-60 minutes from experience.