Intersection (vs. Roundabout)

Problem definition

During the last year almost every day I go around an intersection at Střelnická / Třebenická which has a really unique shape which reminds me of a Roundabout. I'd like to try to simulate the Intersection as an agent based simulation. There are 4 exits from the road: - to Kaufland (biggest local shop, about 130 parking places) - to Ďáblický hřbitov and local housing estate - to Kobylisy - to Střížkov On every exit there are traffic lights with different timings, which I'll easily measure during one measurement at one specific time and day of the week. Most of the cars go straight on the main road from Kobylisy to Střížkov, they also arrive mostly at higher speeds than others leaving from the housing estate and Kaufland. I'd like to propose a simulation based model from NetLogo library with an intersection with 4 exists as is the real example. It will allow users to set frequency, deceleration, acceleration of cars from each direction. The lights will be set to constant ticks which will try to simulate the real scenario. I will omit the fact that cars entering the Kaufland parking place have to leave it also, it will behave as a normal street. Then I'd like to try out a simplified model of Roundabout with the same parameters to compare which of these are better. The simulation would consider some safe distance before another car can enter the roundabout and will have same settings and parameters as intersection. Theoretically I'd love to have two models in one NetLogo project, but I'm yet unsure whether its possible.

Real data

Real measured and approximated frequency of incomming cars are following:

- northbound - 6
- westbound - 48 
- southbound - 6 
- eastbound - 54

Real lights timings measured and approximated are following:

- north-south - 10s 
- east-west - 30s
- yellow - 3s 

Method

The model requires cars entering at different locations, speeds, exiting at different exits and being autonomous in the defined space and environment. The modeling tool should be able to take in account all these facts, so a agent based simulation seems like a clear choice. The other choice could be SimProcess with defined throughputs of each sub-intersection et cetera, but NetLogo is more suited for this kind of modeling and is way more intuitive and the model is almost self-explanatory.

Model

The intersection was simplified to allow easier modeling in NetLogo. The limitations of the model are written in following chapter. The NetLogo world size was set to {{-15,..,15},{-15,..,15}} and in the middle of the model is the intersection. Cars are entering from 4 sides and can exit at any side. All roads were modeled as dual ways as it is in reality, but the cars are unable to change lanes.

The model itself is based on a model Traffic Intersection by U. Wilensky, but it was vastly extended to suit the simulation needs. But the basic settings and description can be found in the base model Traffic Intersection and also in model Info in the NetLogo file.

Although the description and proposal was stating that I'd love to compare this intersection to a roundabout, I was unable to trustworthy simulate the behavior on a roundabout. But in the model there is a scenario that the intersection's lights are down and are all green at once. The cars behavior with lights off is exactly the same as if the lights were on, which is a highly unlikely behavior, because most of the drivers do slow down when they see a problem with lights - e.g. blinking yellow light. Car can turn right (first sub-intersection of the intersection) or left at any sub-intersection, as is complaint with Czech driving laws.

The lights are timed and synchronized in two directions (without priority) to east-west and north-south. The exit direction of each car is generated based on the set probability settings in the model. For exact algorithm see the sub make-new-cars-sub. It can be and on default is disallowed to turn by 180 in the intersection and exit at the incoming direction, because of the model settings limitations which doesn't allow explicitly setting this scenario. If it is allowed there is a high chance in the default scenario that the car will use it with 40% to turn around - this behavior was not perceived during observation.

Limitations

- Crossing and related lights timing was omitted.
- The right turning lane in west-east direction was omitted.
- South exit behaves just like another normal road, it doesn't take in consideration that all the cars exiting in this direction have to exit the same, because it leads only to a parking.
- Cars cannot change their lane if they see the other one is more free.
- Cars enter the intersection even though it seems full and jammed.
- The roundabout was simulated really simply - just by having the lights dead. 

Results

1. The simulation with real measured data is suitable and work in a similar way as the real intersection, except the limitations noted in previous chapter. As we can see the on the screen-shot from NetLogo after over 15 000 ticks the intersection is not jammed and the incoming and outgoing cars from each direction quite represents the reality. 2. If you set the green lights timings without a correlation with the frequency of incoming cars from respective directions, you can see that the intersection jams. If the frequency of incoming cars from each direction was changing rapidly in time, the light timings would have to be adjusted to mitigate the risk of jams. 3. If you kill the lights (dead-lights switch to ON position) you can see that the waiting time of cars from any direction is almost zero and it allows most cars to go through intersection almost without stopping. But I think it is a flaw of the model, because in real-life scenario with lights dead most of the drivers would slow down and some of them could panic which would lead to random stops and maybe even some accidents. Although the simulation propose that if the intersection was converted to a roundabout, it could be more effective (in terms of waiting time of cars). A specific and rigorous simulation would have to be performed to validate this theory.

Conclusion

The problem was to simulate and indicate if it would be better to turn this intersection to a roundabout. The model suggests that the roundabout could be more effective, but the drivers behavior was not simulated properly so it would require more detailed and more complex study. This this simulation vastly extends the current models in NetLogo because it added turning, exiting at different locations, dualways and few other ideas.

Citations

Wilensky, U. (1998). NetLogo Traffic Intersection model. http://ccl.northwestern.edu/netlogo/models/TrafficIntersection. Center for Connected Learning and Computer-Based Modeling, Northwestern Institute on Complex Systems, Northwestern University, Evanston, IL.

Code

Model source code File:Xklit17.nlogo