Delays are a pain. One of the biggest problems we have with public transit is never being certain if the trains are going to run on time. That's partly why transit tends to be much more popular where they do—Japan, say—than where they don't—like the US. (I'd say I show up to Penn Station with a 50/50 expectation that an Amtrak train will have been delayed, and who the hell knows when an L train is going to come.)
I wouldn't have pinpointed Sweden as a nation in desperate need of more efficient trains—I once took the transit into and out of Stockholm, and it seemed pretty top-notch—but they're looking to radically improve their system. A team of data scientists and transit experts in Stockholm have built a prediction model that allows them to visualize the entire commuter train system there, two hours into the future. Naturally, they're using it to make an app, too.
The algorithmic model, which they call "commuter prognosis," and claim is the first automated transit forecasting tool of its kind, gathers data about each train in the system in real time. It then makes automated, continual forecasts about any potential delays the system might experience, and how they'll impact the other trains in the network. It works "like a seismograph," its creators claim: "when a train is not on time the algorithm forecasts the risks of delay effects in the entire network by using historic data," they explained in its announcement.
"We can now forecast disruptions in our service and our traffic control center can prevent the ripple effects that actually cause most delays," said Mikael Lindskog, the communications director at Stockhomståg, Stockholm's commuter train operator.
The company's first step is to translate those forecasts to an app that riders can use to access unprecedentedly accurate information about when their trains will be arriving (they call it a "prognosis alert system"). The second is to use the data to eliminate delays altogether.
"Stockhomståg see big possibilities in this model," Wilhelm Landerholm, the data scientist behind the project, wrote me in an email. "What to get in mind is that if this new model predicts a delay of 15 minutes in 45 minutes they might be able to solve this." In other words, once the train operators know it's coming with enough notice, they may very well be able to address the issue—the deer on the tracks or stalled engine or whatever—and make that delay disappear altogether.
"And if they do, the prediction of a delay is wrong due to the correction," he continued. "The nice thing with this new model is that it will take all information it has at the moment into account, and if there is a change in the information it will automatically give new estimates."
"The next step," Johan Salzmann, a spokesman for the project, told me, "is to integrate the prognosis model with other data sources, such as information of the weight of each vehicle, translating into the number of passengers on each train, to give the operational management (traffic control center) a more effective decision tool." Heavier trains are slower, after all, and more inefficient, and empty trains are no good to anyone; phase two of commuter prognosis will include streamlining the system to optimize user demand. And theoretically, taking unused, unnecessary train trips out the system could help reduce delays by relieving congestion.
But first, there's an app (there's always an app). Apps like HopStop have already helped commuters in the US get better data about their commutes, and made the process more pleasurable and predictable. Know exactly when a train is going to leave, delays included, would be good news for transit riders and operators alike.
"It will be called 'Pendelprognosen'—it basically means commuter prognosis," Salzmann wrote me in an email. "No webby tech web 2.0 name here :)" It will be available in Sweden in the fall.