Dear Jakob, I did some very quick experiments with approach #1, and it definitely helped - almost. I still have a massive traffic jam in one part of the city and also the traffic counts are not matched (at least that is what I decipher from the output log of routeSampler), but at least the jams do not occur in areas that affect my public transport lines. It probably needs some parameter tweaking and maybe a combination of #1 and #2 could help as well. I have got some additional ideas as well, in the moment that I have some time slots to spare, I will write it down.
Thanks for your help! Jan On Fri, 17 May 2024, at 1:15 PM, Jakob Erdmann via sumo-user wrote: > Hi Jan, > so far we didn't have to deal with the problem of guessing traffic for the > side roads based only on observations of the main roads. > > I see two possible approaches here and both try to avoid "excessive" traffic > on the side roads: > > 1. trying to get a "better" initial distribution of random routes and trying > to preserve this distribution during sampling (this means not doing > --optimize full but rather --optimize with an integer limit) > this "better" distribution could come from better informed routing (not > with empty-net travel times as randomTrips does) i.e. by doing a few steps of > iterative user assignment (duaIterate.py) > > 2. adding more constraints by defining virtual detectors that set hard limits > on the traffic that can go through side roads. The problem here is that you > cannot yet define upper bounds distinctively from counts > (https://github.com/eclipse-sumo/sumo/issues/6727) but it might still help if > you set low values for congested side streets. Flowrouter already supports > this kind of general constraint > https://sumo.dlr.de/docs/Tools/Detector.html#restricting_generated_routes on > single edges. > > I'm happy to collaborate on this because I see it as a strategic/crucial > problem in simulation setup. > > best regards, > Jakob > > Am Fr., 17. Mai 2024 um 11:31 Uhr schrieb Jan Přikryl via sumo-user > <[email protected]>: >> Dear all, >> >> my apologies if it has been discussed before, I tried to search the archives >> and did not find anything relevant. >> >> For our simulation, I am pondering the possibility to switch from our old >> manually generated traffic to traffic generated from "observation points", >> in our case traffic counts at intersection detectors. The network we are >> working with represents a substantial part of Pilsen (CZ), it is >> approximately 10 x 10 kilometers in size, and contains quite a number of >> small streets besides the main traffic corridors. The intersection detectors >> are located at the main corridors. >> >> Given the absence of turning rates and other information (I might have an >> outdated O/D info somewhere, but I want to avoid it, if possible), I did >> some experiments with flowrouter.py and routeSampler.py, with rather >> pathetic results, I am afraid: >> >> * flowrouter.py seems to pick up a few random locations in the residential >> areas of the network, and generate traffic that goes from those places into >> the main corridors and then leaves the corridor to some other location; as a >> result, the small streets in residential areas are completely blocked by >> traffic as their capacity cannot handle the traffic volume >> >> * routeSampler.py does a better job but given the necessity of supplying it >> with a set of random routes, it tends to pick up several "most promising" >> routes from a quite large route file (cca 500MB) and while the traffic at >> the main corridors is definitely "nicer", the result is to some extent >> similar to the flowrouter.py, with traffic jams in different residential >> areas that in my opinion are now mostly related to low capacity and >> unregulated / priority intersections -- while the number of vehicle sources >> / destination points increases, the capacity of the network in the side >> streets in not sufficient to accommodate the traffic. >> >> Note that our goal is to have realistic, but not necessarily exact >> replication of traffic in the city. We need to simulate the whole city with >> public transport and we want to determine how the daily traffic affects some >> parts of (trolley/e)bus routes. >> >> I think I am doing something wrong. Is there any additional howto / cookbook >> / manual regarding this besides the standard documentation >> (https://sumo.dlr.de/docs/Demand/Routes_from_Observation_Points.html)? I am >> aware of Michael's paper about routeSampler.py, but it does not help much in >> this context, I am afraid. >> >> I can share the network, the scripts I am using and all the other files, of >> course -- the only problem is that the complete dataset is BIG (230MB >> zipped, 2GB raw). >> >> Thanks! >> >> Jan >> >> -- >> Jan Přikryl >> RICE FEL UWB & CTU FTS >> _______________________________________________ >> sumo-user mailing list >> [email protected] >> To unsubscribe from this list, visit >> https://www.eclipse.org/mailman/listinfo/sumo-user > _______________________________________________ > sumo-user mailing list > [email protected] > To unsubscribe from this list, visit > https://www.eclipse.org/mailman/listinfo/sumo-user > Jan Přikryl RICE FEL UWB & CTU FTS
_______________________________________________ sumo-user mailing list [email protected] To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user
