A program that predicts traffic flow

Imagine a system that would tell you in advance which road you should take to avoid traffic snarl! Well, scientists at Nanyang Technological University (NTU) are working over a program that will guide motorists about road conditions and advise them to take routes to steer clear of traffic jam.
The NTU research team has devised a program that will forecast traffic flow on roads up to one hour in advance.
Last week, the research team presented a paper on its findings at the annual Transportation Systems Conference in Anchorage, Alaska, US.
The conference was organized by the Institute of Electrical and Electronics Engineers, which has more than 400,000 members in more than 160 countries.
The NTU research team’s prediction system is based on a database of previous traffic flows on the roads. Singapore Land Transport Authority divides the road network into segments, and tracks the average speed of cars on each segment every two minute. Each segment is 120m long on average. If the average speed on a road segment slows down over time, it means the segment is becoming congested.
The researchers have compiled two months of data collected in March and April last year into their computer system to create a database of traffic movement. To predict whether a road will have traffic gridlock, the system will go through the past data to find the most similar set of road conditions for the same spot and time of day.
What happens next should be the same on both days, allowing the system to create the prediction.
Dr Justin Dauwels, assistant professor at the NTU School of Electrical and Electronic Engineering, said tests had shown that the system’s predictions were 95 per cent accurate for small roads.
He, however, added that the system did not take into account circumstances like floods, changes in speed limits or the building of new highways.
Researchers in US and China have worked out a similar system. Computer giant IBM tested a system between 2006 and 2007 with results that were 85 per cent accurate up to one hour in advance.

Comments are closed.