Hi Mr. Stelios Krasadakis,

I think that your research is very interesting. To provide some ideas, I need to understand the mechanism of the traffic light management system, by knowing:

1. When will the decision be made (continuously after a certain period of time? how long?)

2. How can the congestion be calculated based on the green light duration?

3. In what case will any road be given a priority?

However, I've tried to search for some related researches. Hope it helps.

Best regards,

Ivan K. Singgih

http://www.sciencedirect.com/science/article/pii/S0952197611000777

http://www.sciencedirect.com/science/article/pii/S1877050916310122

http://www.sciencedirect.com/science/article/pii/S1568494616303556

Hello, Ivan Kristianto Singgih and thank you for your attention.

1. There will be a centralized light( maybe the most important traffic light in intersection according to road weight) for each intersection. Decision about light timings for all 4 lights will be taken after centralized light turns back to green( after one cycle of all lights given a green light).

2. About congestion i'm thinking about RFID's or Car detection from camera with Image processing for counting cars in road.

3. Priority will be fixed. Transportation Engineers will be responsible for giving a value between 0-10. (for example a main street going to the center of the town will take 10, another one going to a centralized location will be given with 8...etc ).

I hope i made it clear...

Best Regards.

It is not clear to me how much background knowlege you have, so this answer is quite general.

The complex task you describe allready has a lot of research and commercial products exist. Some examples for using GA for optimising traffic signal control for planning are TRANSYT or our product CROSSIG. For online optimisation I know of networks PTV-BALANCE or Motion. At the University of Crete you also have a well known expert for this topic - Markos Papageorgiou. It might be helpful to get his lecture documents for more details on traffic signal control and boundery conditions like minimum green times or intergreen times..

The task is in general split up into two steps. First a traffic model to estimate the current and/or future traffic your traffic light control should be optimized on to now the demand. If all cars are detected (e.g. your suggestion to use RFID) and you now their behaviour with a certain time horizion an estimation is not needed, for sure.

Second an optimization of your traffic light control for that traffic demand. The optimisation for traffic is normally to minimize delay, stops or nowadays also emissions.

Simple models to calculate e.g. delay for singel intersections from a given demand (the arrivals of your cars at the stop lines) and a given service rate (green times at this stop lines) are from Webster or Kimber/Hollis. For networks Robertsons platoon dispersion model or the Cell Transmission Model by Daganzo are used.

For the traffic light control you have several degrees of freedom like green split, cycle time and stage sequence (a stage a combination of a green light for the non conflicting traffic streams while the others are red) and for network control also the offset.

Now you can generate solutions using a GA and evaluate them according to its delay.

CHROMOSOME DESIGN:

• I think that you may design the chromosome, consisting of 4 genes. Each gene represents the green light duration for each light. You may use the continuous genetic algorithm (GA) method. In the attached file, you may see how the crossover and mutation operations can be performed in this continuous GA.
• I suggest that the road priority can be considered when mutation operators are applied to some chromosomes. You may develop a method to modify the genes in the chromosome based on the given priorities.

CHROMOSOME EVALUATION:

• USING A CONGESTION CALCULATION FORMULA: In the general GA, objective values of each solution (in a chromosome) is calculated using a certain formula. If it is necessary in your case, I am thinking that you may use some traffic congestion measurement formula (I am not familiar yet with these formulas), perhaps combined with a queuing approach, given the arrival rate of vehicles and the service time, which is the time required to pass the intersection due to the congestion.
• USING SIMULATION SOFTWARE: You may model the movements of the vehicles (including acceleration and deceleration rates, maximum speed, required distance between vehicles, etc) in a simulation software. You may run the simulation several times in order to obtained the long-run average number of cars on the road, then use the obtained value as the objective value of the chromosome. However, this approach may take too long time.
• USING HISTORICAL DATA: You may collect historical data from the RFID or camera with image processing to obtain number of cars on the road given some duration of green lights. If not all data exist, perhaps you may use a forecasting method to estimate the number of cars, given a shorter and longer green light duration times.

CONSIDERING VARIOUS INTER ARRIVAL TIMES OF VEHICLES:

• In practice, the inter arrival times of vehicles to a certain destination are different between certain time periods, e.g. there will be less arriving cars at night and more cars right before working time and after working time.
• I suggest that the a certain GA must be performed in each time period, in order to obtain the best decision, suitable with the real situation in during that time period.

Don't forget to reach out to the authors of this research:

Chapter Synchronization of traffic signals through a heuristic-modif...