sardonick

Been wanting to do this for a while:

sigalert.com has wonderful data that can be mined, with regards to patterns in freeway traffic. It’d be great to harvest and analyze traffic trends.

TODO:

• email Ken (urban planning/transportation PhD at UCI) to see if they have any data i can steal
• email sigalert to see if they can give me access to their raw data (probably not… or should i just covertly spider it?)

Pages to gather from:

• http://www.sigalert.com/speeds.asp?Region=Greater+Los+Angeles&Road=405%20South
• http://www.sigalert.com/speeds.asp?Region=Greater+Los+Angeles&Road=405%20North
• http://www.sigalert.com/speeds.asp?Region=Greater+Los+Angeles&Road=10%20East
• http://www.sigalert.com/speeds.asp?Region=Greater+Los+Angeles&Road=10%20West
• http://www.sigalert.com/speeds.asp?Region=Greater+Los+Angeles&Road=5%20North
• http://www.sigalert.com/speeds.asp?Region=Greater+Los+Angeles&Road=5%20South
• (and perhaps others…though don’t want to overload their servers too much or alert them to my presence)

Programming TODO:

1. figure out how to use cron and wget
2. start sucking down pages, every 25 minutes or so
3. write a parser to grab the necessary data from the pages in question
4. find a good DB backend to store my data in
5. cron extract freeway speeds and write them to DB (get it all working together)
6. data processing
   • current speeds
   • average speeds (model this after web-traffic data: avg per day-of-week, month, weekday/weekend, etc)
7. figure out distance measures between points on the map
8. use the prior to calculate a numerical value of travel-time (e.g. “25 minutes if you leave now”)
9. explore equations that fit traffic
   • accidents
     • measure severity by time-to-normal
     • clustering of severity
     • resolution with respect to location (because the front cars are accelerating and the back cars are slowing down, congestion would move backwards along a road like a wave, even once the original accident site is resolved, huh?)
     • resolution with respect to time (how long does it take an accident to dissolve?)
   • rush-hour trends (can i treat rush-hour and congestion as an accident without a specific center?)
   • smoothing, with respect to day-of-week, accident resolution, hour-of-day
   • The effects of reverse commutes (e.g. the 405 where both sides are backed up)
   • Average traffic speeds for specific exits
10. adapt estimated time with averages, to see if I can get an estimate of “how long will it take me if i leave +5 minutes, +10 minutes, +20 minutes, etc…”
11. measure a minutes-here-waiting-vs.-minutes-on-the-road ratio
12. put all of this into a web frontend

Leonard recommends

• This for an example of www:mechanize and tokeparse
• DBI to hook perl into mysql
• output to web using php linking into mysql
• jpgraph as an alternative to gnuplot which i was thinking of using

Ken says:

• The Transportation Research Board has a bunch of good papers
• The University of California Transportation
  Center does too, specifically
  • http://www.uctc.net/access/access.asp
• But, analyzing traffic dynamics just by knowing traffic speeds at points on the freeway is a pipedream…there are way too many other variables involved.
• UCI’s ITS (and UCLA, too) should have all the data i could ever want…
  • Brian Taylor is a “transportation guru” at UCLA
  • Hiro Iseki is a PhD candidate at UCLA (hiseki @ ucla.edu) (his background is in transportation engineering )