Check counts for switzerland

[sebhoerl]

• [x] Run scenario synthesis with count matching location assignment
• [x] Validate with existing count analysis tools

[ctchervenkov]

So I made a first attempt at looking at the counts for Switzerland. street counts : https://repos2.ivt.ethz.ch/svn/ivt_studies/trunk/baseline2010/supply/base_street_counts/streetCounts_Hourly.csv network : https://repos2.ivt.ethz.ch/svn/ivt_studies/trunk/baseline2010/supply/base_multimodal_supply/CH1903Plus_HAFAS2015/mmNetwork.xml.gz events : 20.events.xml from a 10% Swiss scenario we were using with Joe for externalities

Then using the analysis tools in the baseline scenario repo, I compared the hourly counts. Multiplying the simulated counts by a factor 10, this is what the comparison looks like:

In other words, NOT GOOD! Next question is how do we fix it.

[sebhoerl]

Ok, so probably the events don't fit to the network / counts anymore. Next steps:

• [ ] Generate new Switzerland scenario and run it
• [ ] Take either the mentioned counts (or even the raw data) and map them again to the new network (standard approach would be by matching centroids of the links and angle, there is probably already a script for that in the repository)
• [ ] Compare again with new matched counts and new events file

[sebhoerl]

But if it looks like this then it probably means that we distribute the vehicles to much over the network. So actually increasing the flow capacity might make sense, or adjusting the speeds.

[sebhoerl]

And would be interesting to see this separate for the counts of Kanton / City of Zurich / ASTRA Hgihways.

[joemolloy]

well it looks like the magnitude is way off. what about as a start comparing the total counts from the reference and scenario*factor?

[ctchervenkov]

Well, again from the results of the analysis tools that computes the hourly reference and simulation counts, I have a total of 74,067 simulated counts versus 13,121,268 reference counts, which is a factor of 177 (so with the factor=10 then that gives 17.7). Something indeed looks fishy...

[sebhoerl]

AGAIN: We have neither cross-border traffic nor freight traffic and we lose some of the agents during the scenario synthesis (e.g. those which do not start and end a plan at "home"). So there is no reason why we should have a factor of 10.

Just find the scaling factor that minimizes some distance, e.g. L2 norm over the difference in the count stations. What matters more is the relation between the count stations, that's why I usually displayed them ordered by reference count. This will show you if "large" count stations also produce "large" counts in the simulation and the same for "small" ones.

[sebhoerl]

And have you done the checks yet whether the mapping of the count stations to the links is correct? Before that there is no use in comparing them ;)