Goal
The objective was to determine key metrics to understand what type of stop/traffic lights is best to use. By creating a stochastic simulation based on real world data we simulated an intersection with low, mid, and high traffic volumes to compare the different types of stops:
- Two-way stop
- Four-way stop
- Four-way traffic lights (actuated – sensors)
- Four-way traffic lights (timed)
The data analysis was done in R, Excel, and simulated in Arena (discrete simulation), created by Fabio Santos and Audrey Lai.
Analysis
Data was provided by the Govt. of Canada, 2020, Montreal. Data wrangling was done in R and Excel, producing average traffic flow from four direction (Northbound, Eastbound, Westbound, Southbound) at a 15 minute intervals.
Some notable changes to the original dataset:
- The traffic patterns are different between weekdays and weekends. To parse the patterns, our scope was limited to weekdays, so weekend was filtered out
- Pedestrian were considered a vehicle type. As pedestrian travel at a significant slower speed than other vehicles, and how they affect lights, pedestrian were filtered out
- Traffic flow is defined as the sum of vehicle arrivals from four directions. The cleansed data was aggregated into three segments of traffic:
- Low volume
- Mid volume
- High volume
It becomes very clear that there is an Eastbound peak and a Westbound peak at different times of day, probably going and returing from work, or a similar pattern. To continue with the simulation, we extracted the probailities of which directions the cars would turn over the entire period:
We can now assume that this intersection most likely does not allow for a North direction. Since 0% go NorthBound Straight, or turn left when EB, or right when WB. Most likely being a T shape.
Simulation
We used Arena to simulate a Two way stop (TWS) and a Four way Stop (FWS), and made sure the model worked whether there was inbound traffic from all directions or not. This would help us later on, as we could try various types of distributions for the cars and confirm our findings.
Then we set each inbound direction with the proper distributions over a 24 hour period, this allowed us to get what a single “average” day of an intersection produced in results. We primarily measured the wait time and the number of entities that moved to and from directions.
Then we repeated this simulation with a four way light actuated system. Where the lights would only change if there were cars waiting to cross. Then similarly with a light timer system, which switches for red and green lights on a simple timer.
This was also made to interpret results from all diections and we would allow the distributions themselves to sort any ‘north incoming’ cars, as those were 0.
Conclusions
We found that the most effective models (unsurprisingly) were a properly set Two-Way-Stop and a Four-Way-Stop with actuated lights. Although some caveats should be noted.
We looked mostly for: traffic flow (how many cars go in a given direction), average wait time, and long wait times. We particularly noticed how sometimes an actuated system could have the high wait times depending on its configureation.
- Two-way-stop on EW side: while average wait time increased as traffic flow increased, the setting demonstrated the shortest average wait time. This only works if the stop is on “south” part, and severly hurts incoming south turning left
- Four-way-stop: average wait time was higher than that of two-way stop, and increased exponentially as traffic flow increased
- Four-way-light (actuated): saw the longest average wait time due to lack of timeout limit. If timeout limit is built in, this could potentially be the shortest wait of the four scenarios given any traffic volume
- Four-way-light (timed): while average wait times were longer than two-way-stop and four-waystop, it was the most efficient in avoiding long wait times