The traffic situation at urban expressway interchanges is really complicated in
daily life. Cars change lanes very often, and problems from cars merging
together are obvious. Traditional traffic models aren’t accurate enough when
they try to predict what happens in these areas. To solve this, we suggest a
better cellular transport model (CTM) that’s improved using genetic algorithms.
It can describe and improve traffic conditions in a flexible way.
We picked the interchange on Hohhot’s North Second Ring Expressway for our study.
To get traffic data during rush hours—7 to 9 in the morning and 5 to 7 in the
evening—we used a few methods together. There was video monitoring with tools
like YOLOv8 and DeepSORT, people counting cars by hand, and also VISSIM
simulation. The data we collected had things like how fast cars were going, how
many were packed in an area, and how much traffic was moving through. With this
info, we could see how traffic changes in different parts of the interchange and
at different times.
Traditional CTMs have their limits. The cells in them are stuck at the same
length, their capacity never shifts, and ramps are updated the same way every
time. So we fixed three things to make it better. We made the cell lengths
change based on how heavy the traffic is. In areas where cars move freely, the
cells are split into bigger chunks. But in busy interchange spots, they’re
divided into smaller, more detailed pieces. We built a capacity model that can
adjust. It uses something called a “bottleneck coefficient” to figure out how
much capacity drops when cars merge and cause issues. By mixing virtual cells
with real ones, the capacity of the ramp to handle the traffic is improved. This
enabled the model to show how waiting in queues affects the number of cars on
the road.
Validation results show that compared with the traditional model, whose MAPE is
12.3%, the improved model has a mean absolute percentage error (MAPE) of 5.89%.
Its root mean square error (RMSE) is 28.6 vehicles per hour. For the traditional
model, this number is 67.2 vehicles per hour. So, the accuracy has improved by
57.4%.
When we used this improved model to test a new exit plan, the results showed
positive signs. During peak hours, the road’s capacity could go up by 16.75% in
the morning and 26.02% in the evening. At the same time, serious traffic
problems would drop by 52.51%. This shows that the better model can really help
make good decisions when optimizing busy traffic areas where cars cross each
other.