meso-tls-penalty underestimates effects from queuing

[namdre]

The current code estimates the average time until the light switches to green and uses this as penalty.
With active junction control another effect is observed: vehicles have to wait until the queue ahead of them clears and they can move up to the stop line before actually passing the traffic light.

[aminissn]

@ambuehll

[ambuehll]

I thought this was an expected behaviour based on the discussion in #5746.

[namdre]

Yes (https://sumo.dlr.de/docs/Simulation/Meso.html#tls-penalty).
The current computation gives a lower bound which is correct for very low levels of traffic.
At very high levels of traffic, the meso-tls-flow-penalty serves to correct the throughput to maximum capacity.
However, at the level of traffic where flow is just below capacity (no upstream queue buildup), the lower bound may be quite far away from the delays that would be realistic.

This ticket serves as a reminder to improve the penalty computation or at least document the mismatch (and suggest that users switch to meso-junction-control if they need the accuracy)