Re: [sumo-user] Realistic Simulation with traffic data and calibration

[Jakob Erdmann <namdre.sumo@xxxxxxxxx>]

Regarding jams: I don't think that there is a traffic-count mismatch. After all, the traffic counts at the intersection should be no higher than you real-world measurements.

Also, by using turn-counts instead of edge-counts, a big part of the route-related ambiguity is eliminated. At least when taken in isolation, every intersection should be a close match to reality.

From my experience, there are plenty of things that can go wrong with turn lanes and lane-to-lane connections that can reduce intersection flow and create jams in a scenario that is already close to saturation.

Here is a somewhat systematic approach to understanding the jams:

- pick one of your intersections with lots of traffic that appears to be oversaturated (queues keep growing over multiple traffic light cycles)

- create a sub-network with just this intersection and plenty of incoming edge length (to better observe growing queues)

- use routeSampler to create traffic for just this intersection (it should be easy to get a perfect traffic match for a single intersection)

- look at the simulation carefully,

  - Are all lanes used?

  - Is the traffic light switching reasonably?

  - Is the measured flow even plausible?

  - Are there any deadlocks?

If you still cannot figure it out, send me an offending single-junction scenario at I'll check it out myself.

regards,

Jakob

Am So., 12. Apr. 2020 um 22:43 Uhr schrieb Jakob Erdmann <namdre.sumo@xxxxxxxxx>:

regarding public transport for a reduced network, see https://sumo.dlr.de/docs/Tools/Routes.html#cutroutespy.

Am Do., 9. Apr. 2020 um 06:55 Uhr schrieb Tetris <schmelter@xxxxxxxxxxxxxxxxxxx>:

Hello,

I'm still figuring out what causes the traffic jam in the simulation. There
is no specific junction at which the traffic jam starts. It's more in
general that there are way too many vehicles.

I'm going through the list from
https://sumo.dlr.de/docs/FAQ.html#the_simulation_has_lots_of_jamsdeadlocks_what_can_i_do
:

1. invalid network

    invalid lane numbers -> where can I see that the lane numbers are
invalid? netcheck.py? But netcheck.py is for edges, not lanes.

    missing turning lanes -> Does this mean turning lanes at junctions? This
is something I could check just by looking at my network. I created routes
for every turn for all the junctions in my network (at least for those that
have traffic counts)

    invalid connections -> that's checked by netcheck.py?

    invalid junctions (big clusters of small junctions should be joined) ->
used join junction command.

2. invalid traffic lights (see Improving generated traffic lights) -> I used
actuated traffic lights.

3. invalid demand (too many vehicles overall, too many vehicles starting on
the same edge). -> This seems to be causing the trouble.
I haven't defined starting edges, just the routes and maybe too many routes
start with the same edge?
The traffic demand is build upon traffic counts which have been made on
different days, even different months. Nonetheless I just put these traffic
counts into the turncounts.xml for routesampler.py. Can I calibrate the
traffic counts with Cadyts or is Cadyts for different use cases?
https://sumo.dlr.de/docs/Contributed/Cadyts.html

4. invalid routing
    only shortest path were used instead of a user assignment algorithm -> I
used randomtrips.py as input for routesampler. Would it be useful to run
duaIterate.py on the random trips?

    too many vehicles start/end their route with a turn-around. -> I have
seen in the simulation that vehicles turn around at the end of the road
where the network ends.
https://sumo.dlr.de/docs/Simulation/Turnarounds.html#controlling_where_turnarounds_may_happen
I could use one of the --no-turnarounds options in netconvert.



5. invalid insertion (vehicles being inserted on the wrong lane close to the
end of an edge where they need to change to another turn lane). This can be
fixed by setting the vehicle attribute departLane="best" -> I already
updated my rou.xml and inserted departLane="best" to every vehicle.

I think the main cause of the traffic jam are the traffic counts that don't
fit together.
https://elib.dlr.de/127994/1/08569938.pdf The chapter V-B"demand adaption"
suggests 3 methods: Cadyts, SUMO Calibrators and REROUTER.
Which of those 3 methods fits the best to routesampler.py output from random
trips? 





--
Sent from: http://sumo-user-mailing-list.90755.n8.nabble.com/
_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user