Dear Jakob,
thank you very much for your answer!
I made an experiment with 100 replications storing the values of totalTraveTime and totalDepartDelay, using the same base demand with a little randomness, always with the base signal timing, without using Webster, just to check the effect of the intersection model. The results show that the ratio totalDepartDelay/totalTraveTime is in average 0.762 (76.2%) with 0.192 of standard deviation. That would mean that I have virtual queues? But I have watched some of the simulations using the GUI and visually the queues by far do not reach the entry points of the model during the all 3600 seconds of the scenario.
Regards,
Alexandre
---
Prof. Alexandre Hering Coelho, Dr.-Ing.
Departamento de Engenharia Civil
Universidade Federal de Santa Catarina
Florianópolis, Brasil
Hello, I would generally expect that modifying cycle times and green split according to Webster leads to performance improvements compared to unoptimized fixed plans. Could it be that your scenario has significant departDelay in some of your runs? A likely reason for unexpected performance measures is that one scenario has much departDelay and (comparatively) less timeLoss (since timeLoss is only counted after departure) while another scenario that works better, has less departDelay but more timeLoss. A simple way to check is by using statistic-output and comparing the sum of totalTraveTime and totatDepartDelay (for scenarios that have the same demand input and run to completion).
regards, Jakob
Hi.
I modeled a simple 4-legs intersection with traffic lights. With the help of Python I gradually increase the demand volumes, with a certain random margin. At each iteration initially I set the cycle time equal to 60 seconds, the yellow times equal to 3 seconds and the green times equal to 27 seconds, I use Webster's method to determine the optimal cycle time and the green times and compare performance metrics between the two scenarios (edge based, statistics and queues). I'm very intrigued, because most of the time the performance is much worse after optimizing the times. Assuming that my calculations are correct and that the logic is all correctly implemented (I have reviewed everything dozens of times for days), could anyone give any clue to the reason for this strange behavior? Or is it always expected that performance will be better after applying Webster's method, even at microscopic scenarios, and so can only be a problem in my implementation?
I've already experimented with different models of car following, with different values of reaction time (action-step-length) and with different values of lost times (Webster) and nothing helped.
Thank you.
Alexandre
---
Prof. Alexandre Hering Coelho, Dr.-Ing.
Departamento de Engenharia Civil
Universidade Federal de Santa Catarina
Florianópolis, Brasil
_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user
_______________________________________________ sumo-user mailing list sumo-user@xxxxxxxxxxxTo unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user
|
