Traffic and cascading information

People

Giovanni Petri
Leif Karlstrom
Drew Levin
Tracey McDole
Samuel Scarpino
Kang Zhao

Intro

Transportation networks represent an element of enormous importance in the development of economies and in our daily lives. However, the common experience of such networks is often very poor due to congestion and scarce available information. It becomes important than to consider what are the effects of real-time information about the status of the network itself. Not much work has been devoted to this subject, but this is becoming more and more relevant as our capacity of sensing and communicating are growing very fast. We want to consider modifications of this model, where decentralised information dissemination is used to study the emerging length scale of interaction.

Questions

(1) What happens when one specifies the model better? What if the network is driven, e.g. a net flow through the network ?
More specifically, when one introduces a specific origin-destination table, does it change the performance of the system? If yes, where does the reduction in performances come from? If no, why does it still persist? Is it enough in this frame to consider cars as (almost) randomly diffusing?
(2) Is it possible to let the parameter commanding communications evolve over time copying neighbours dynamics.
(3) the original model was cast as a combination of a ZRP process with a cascading process on top. Both processes are (fairly) understood, it would be interesting to see whether we can come up with a (tentative) analytical description. Anyone feeling like it? (4) what are the effects of taking into account the information spreading? (5) how does it relate to the memory effect in my other model? Does it make sense to include a memory in the information spreading? (6) is there a transition? or the effect is smooth? (7) what are distributions and shapes of avalanches? is it possible to exploit them to control the system?

Timetable

17-18 getting familiar with code, modifications and first simulations
19-20 realistic maps (maybe), launch of set simulations on different topologies
21-23 refinement, additional simulations, interpretation
23-24 poster preparation and final touches
25 be done with it (for now at least)

Stuff

You should find the code in your mail, if not call/mail me. We are getting maps from the S&T team possibly by friday or so. The code has many tunable parameters, that must be carefully chosen to be able to interpret the results with any confidence. There are comments in the file, however I fear they are nowhere near being satisfactory so for any problems just ask me straight away. Also, if you want to change/edit/trash it, feel free.

- first, there's alpha, power of the distance, involved in the navigation probabilities' calculations. That has to be chosen as to provide reasonable travel time over test distances in the limit of zero density (one car at a time traveling, so no interaction with other cars). Right now, i set alpha=5, it seems to be working pretty well, but maybe 3-4 could be as good. Needs to be checked.

- second, once this parameter is set, we need to have a reference forcing for the system, which translates to how many cars I'm putting in at any time. My suggestion is to test putting higher and higher forcing in a system where the information spreading is switched off (can be achieved by simply setting the compliance of the travelers to 0). Doing this allows to find the maximal forcing that the system can support without developing severe congestion.

- once the maximal forcing is identified, it is interesting to switch on the interaction (information spreading), studying what changes as the threshold for signaling varies.

-if an optimum is found (say for threshold 0.3 the system looks stationary and delivers supermany more cars than in the other condition) it becomes interesting to study how resistant is the system at that point when the forcing is increased even more (increase the driving until something SPECTACULAR happens!).

- The eventual natural extension in my opinion would be studying a bit more the parameter space, for example studying how/whether the optimal signaling threshold depends on the forcing, topology, compliance of drivers..

-leaf mentioned being interested in seeing whether it is possible to create a natural main flowpath by manipulating the capacity/signaling of parts of the system. I think it's nice, and it would be nice also to see how this behaves when congestion starts developing on the main flowpath and how it spills into the "secondary" paths.

Code Output Files

the (messy) output files the codes is spitting out are:

Total population as function of time
Total delivered cars as function of time.
Travel time of delivered cars
file containing travel time and length of shortest path between origin and destination for each delivered car.
Populations of nodes (obtained as the sum of the populations on the links incoming to the node).
Paths taken by a selection of cars (to be implemented)
Size of cascades of informations
Queue distributions as function of time

Ideas/Discussion

The code for traffic simulations with lattice, random graphs and BA graphs created internally can be downloaded from here.

I'll put on the code usable to import maps that sam made.