Bad driving holds the secret to traffic forecasts

A traffic simulation system is helping drivers by predicting jams on Germany’s autobahn network up to an hour before they happen. The secret of its success is to take into account the way real drivers – and their cars – behave. When engineers model the way road traffic flows they break the traffic down into three categories: freely flowing, jammed, and an intermediate state called synchronised flow in which dense traffic moves in unison, like marchers moving in step.

But this synchronised flow is unstable. One car pulling into another lane and forcing the driver behind to brake hard is enough to start traffic bunching up. This can quickly develop into a jam that propagates backwards through the traffic like a wave. Failure to predict this “pinch effect” has stymied past attempts to model traffic flow.

Now Michael Schreckenberg and colleagues at the University of Duisburg-Essen in Germany have developed a computer model that successfully reproduces the pinch effect. “It is the first model to reproduce all known traffic states,” says team member Robert Barlovic. The team’s trick is to be realistic about driver behaviour. “Real drivers tend to hinder each other when doing things like changing lanes. All this has to be taken into account,” says Schreckenberg. And where previous models have simplified the way cars move- by assuming they can stop immediately without slowing down first, for example- the new model is more sophisticated.

Schreckenberg’s computer model divides the road into a regular grid, with one line of cells representing each lane on a highway. Cells in the grid are marked as either containing a vehicle or empty. The number of empty cells between the virtual vehicles depends on the way the drivers are behaving. Accuracy not seen before has been achieved by modelling two behaviours, says Schreckenberg. These are dubbed “aggressive” behaviour, in which drivers either get too close to the car in front and have to brake, or in which they change lanes too quickly, forcing others to brake. The second behaviour is “defensive”, in which they drive at a generally safe distance.

As the model runs, it moves vehicles according to rules that embody realistic rates of acceleration and deceleration. No infinite decelerations are allowed, for instance. The result is a software model that combines realistic driver behaviour with realistic physics.

The model is already being used to forecast traffic on the autobahn network around the city of Cologne, based on traffic data gathered in real time from sensors buried in the road. Its forecasts, which predict conditions up to an hour ahead, are displayed on the web at www.autobahn.nrw.de. More than 90 per cent of time, it correctly predicts traffic density.

But the website has already become a victim of its own success, admits Schreckenberg. Some of the 300,000 people a day who are visiting the site are replanning their journeys on the basis of its forecasts, and this is beginning to make the forecasts themselves less accurate. And soon it could get even worse when the website becomes available on 3G cellphones, he says.

So the researchers are now trying to adjust the way the traffic information is provided to drivers to take this destructive effect into account. One idea, says Schreckenberg, might actually be to provide less complete traffic information to encourage drivers to adopt more varied strategies for evading congestion, so they do not all flock to the same exits.