Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [sumo-user] Different behavior between hourly simulations and a single 24h SUMO run (flowrouter + calibrators)
The edge-count data input files for flowrouter and routeSampler are both able to describe time-varying curves and are freely convertible between each other
with edgeDataFromFlow.py and flowFromEdgeData.py (https://github.com/eclipse-sumo/sumo/issues/6609)

For large networks it is generally advisable to perform longer simulations to avoid warm-up related artifacts.

Am Fr., 9. Jan. 2026 um 13:15 Uhr schrieb Pedro Oliveira <pedro.engenhariae@xxxxxxxxx>:
Good morning! For 24-hour traffic, wouldn't it be interesting to distribute the traffic based on the distribution curve throughout the day? I don't know how the flowrouter file that loads the traffic is written because I use routesampler. In routesampler, for example, the volume is described within a range: 0 to 3600, 3600 to 7200, etc. So, based on the volume at each time, it's possible to insert the desired volume. Couldn't the same procedure be done with flowrouter?

Sds,

Pedro Oliveira 

Em sex, 9 de jan de 2026 05:53, Jakob Erdmann via sumo-user <sumo-user@xxxxxxxxxxx> escreveu:
In the 1-hour simulation you are always starting with an empty network and thus large amounts of vehicles are absorbed by the empty road network without causing congestion.
In the 24-hour simulation, most vehicles are released into a filled network and thus any mismatch between traffic model assumptions and network capacities may easily turn into congestion.
You will need to do the hard work of figuring out where jamming starts and whether it comes from problems with the traffic, network or infrastructure.

If your counting data is sparse in space, the solution space encompasses solutions with more or less total traffic (in the extreme cases, vehicles could either start just before a detector and arrive directly after it for a minimal-traffic solution or they could take large detours through roads not covered by measurement data for a maximum-traffic solution). 
If you notice problems with too much traffic on edges where detectors are lacking you could set option --max-flow and --max-turn-flow or load --flow-restrictions for such edges to improve the flowrouter solution.

Flowrouter defines flows with equidistant spacing in time (using 'begin', 'end' and 'number': so it already is as smooth as possible.

regards,
Jakob

Am Do., 8. Jan. 2026 um 17:33 Uhr schrieb Gabriel García Casa via sumo-user <sumo-user@xxxxxxxxxxx>:

Hi all, I hope you had a pleasant Christmas holiday.
As said in previous posts,  I am working on a calibrated SUMO model where traffic demand and speeds are derived from real measurements.

My workflow is the following:

  • Traffic counts are matched using flowrouter.py with hourly counts (interval = 60 min).

  • Route feasibility is enforced using implausibleRoutes.py, and flowrouter.py is run again with the resulting restrictions.

  • Speeds are calibrated using speed-only calibrators, with one speed profile per edge and per hour in the traffic measurement points (e1 detectors).

What I observe is a strong difference depending on how the simulation is run:

  • If I run independent simulations of 1 hour each, the model behaves well and congestion levels look realistic.

  • If I run a single continuous 24-hour simulation (0–86400 s) using exactly the same routes, flows and calibrators, the network progressively builds up heavy congestion and ends with many vehicles still running or waiting.

In particular, I would like to understand:

  • Whether flowrouter.py tends to insert vehicles in a concentrated way at the beginning of each interval when using hourly counts.

  • If there is a recommended way to smooth or distribute vehicle insertion over time when working with coarse (hourly) demand. I was thinking about programming a smoothing function.

  • Whether the combination of route restrictions and speed-only calibrators can unintentionally reduce effective capacity over time, leading to queue accumulation in long runs.

  • More generally, whether this behavior is expected in SUMO and how such effects are usually mitigated in full-day simulations.

Any insight or pointers to relevant SUMO options or best practices would be very helpful.

Thanks in advance.

Gabriel


_______________________________________________
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@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user

Back to the top