Hello,
I have been trying to replicate the experiments form the paper:
"
Proposing a
simulation-based Dynamic System optimal traffic assignment algorithm for SUMO
an approximation of marginal travel time"
I was able to get the same random network from the authors. I then randomly generated 7200 vehicles over a 3600 second time period, and then used duaIterate's SUE and SDSO method alongside a deterministic DUE and DSO method I made myself. (the probabilities of the routes with the lowest cost is set to 1 in the rou.alt.xml file). I also vary the use of Logit.
I repeat the above tests 10 times, then select the iteration in which each method produces the lowest average travel time and aggregate the results per method to get the following graph:
While this configuration mostly follows what the paper mentioned (DSO doing better than DUE, SDSO doing better than SUE, and SDSO_logt doing better than all other LOGIT methods), there are a few issues I've been trying to resolve.
First, the methods not labeled "LOGIT" are using the gawron route choice algorithm, and from all my experiments it seems that Gawron always gives better performance than LOGIT. I was wondering if this has to do with the fact that I'm considering a Microscopic simulation or if there was some other reason, such as incorrect configuration on my end.
Second, If I vary the numbers slightly, such as considering 6400 vehicles I get the following:
Where now both SDSO methods are being out performed by their relevant SUE methods. The same happens with 10,000 vehicles and also happens if I change the map.
My issue is that the DSO results seem to be rather unstable and I was hoping to see if this was something on my end or more of a product of using an approximation of DSO.
I should also mention that I have tried using the PSWAP method instead of MSA and have rather negligible performance improvements if any, again I have no idea if configuration is an issue.
I hope that this is the right place for this, I have been struggling with this for a while and was hoping to get some help/feedback. Please let me know if anything seems unclear and I can make clarifications.
Best regards,
Thomas Carroll
