And now, here is the traffic forecast ...

 作者:呼延叼     |      日期:2019-03-01 05:03:05
By CHARLES ARTHUR COMPUTERS may soon be able to predict road traffic patterns across Britain, just as they forecast the weather, as a result of a project running on Europe’s fastest computer at the University of Edinburgh. The computer can simulate the individual routes and driving behaviour of 250 000 vehicles as they travel around Scotland’s trunk road network. Because this number of vehicles exceeds the present rush hour figure and the software can run the simulation five times faster than real life, researchers believe that given real roadside data, the model will produce accurate forecasts. This could be used to advise people how to dodge traffic jams, just as weather forecasters can predict the arrival of rain. The lifelike results of the most recent simulations have surprised even the researchers, a joint team from the university and SIAS, a transport planning consultancy based in Edinburgh. For instance, the computer model produced both the “shock waves” of congestion that travel backwards down motorways and create traffic jams where there is no obvious obstruction (“When shock waves hit traffic”, New Scientist, 25 June 1994), and the “slow fast-lane” effect, in which so many drivers move into the overtaking lane in frustration at the middle lane’s lower speed that the middle lane becomes the fastest-moving. The software works by modelling the action of individual vehicles, each of which belongs to one of 14 types (such as cars, trucks and buses) with 330 different origins and destinations. The system then sets them all loose on Scotland’s trunk roads, using detailed maps provided by the Scottish Office, and follows the effects, both at a microscopic and macroscopic level. This method is a radical change from most road network modelling systems, which treat traffic as if it was an ideal fluid passing along a system of pipes. Such a model cannot include the interaction of individual vehicles either at junctions or on the open road. In fact, though the new method eventually arrives at a model of fluid flow, it has done so by the much more accurate process of following the actions of individual molecules. “That is the crux,” says Gordon Duncan, who is leading the group at the university’s parallel computing centre. “Physical geometry, traffic control signals, vehicle type and time of day all make a significant contribution to traffic flow at junctions. Unavoidably, the flow there affects the roads that lead up to them. A truly accurate picture of the expected pattern of congestion is impossible without microscopic simulation.” The joint team has been working on the project for two and a half years. But the breakthroughs have come late. The most surprising aspect, says Stephen Druitt, a SIAS director, is how easy it is to model the behaviour of drivers on motorways, whose decisions can be reduced to three variables: the distance to the car in front; the distance to the car behind; and the presence or absence of a gap in an adjacent lane. When vehicles are given a range of speeds based on a normal distribution but allowing for road speed limits and differences in top speeds, real-life situations begin to crop up on the simulation. “In the last fortnight we have duplicated spontaneous congestion on motorways,” says Druitt. “Now we have to prove that the model works for real data.” The data could be gathered from roadside monitors, such as cameras on motorway bridges. Druitt believes a pilot study using real data will demonstrate the validity of their model within the next 12 months. Mark Smith, head of the business systems group at the computing centre, says the team’s medium-term goal is to give forecasts of traffic patterns up to 20 minutes ahead telling drivers how to avoid impending congestion, for example from displays on motorways. “This weather forecasting idea has been at the heart of what we have been going towards for the past year,” he says. “We had funding from the Department of Transport. Usually they fund things like what sort of concrete to put on roads or what sort of iron to use for manholes, so this is very different. But they are being very supportive.” The software uses up to 32 processors on the 256-processor Cray T3D computer at the parallel computing centre. Each processor acts essentially on its own, modelling the behaviour of the vehicles within its own region. When a vehicle “leaves” a region, its details are sent as a message packet to the processor it is joining. This means that the system does not have an overall controller, though the output from the processors can be viewed together. The Cray cost £8 million, making its use for road forecasting alone an expensive proposition, even though a large-scale project using all the processors could model up to 2 million vehicles. To cope with the need to find information at the microscopic level, the team has devised a graphics interface which runs on a single Silicon Graphics workstation, costing about £7000. This can simulate 1000 vehicles in real time on a test network of about 50 junctions. “We don’t need the full power of the Cray to do the really useful work,” says Druitt. “With the parameters that we’re using when we run the simulation,