PPT - The 3rd Conference on Innovations in Travel Modeling
PPT - The 3rd Conference on Innovations in Travel Modeling
PPT - The 3rd Conference on Innovations in Travel Modeling
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
A GPS-based Bicycle Route Choice<br />
Model for San Francisco, California<br />
Jeff Hood<br />
University of California, Berkeley<br />
Billy Charlt<strong>on</strong>, Elizabeth Sall, Michael Schwartz<br />
San Francisco County Transportati<strong>on</strong> Authority<br />
Matt Paul<br />
MoPimp Producti<strong>on</strong>s<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY<br />
3 rd <str<strong>on</strong>g>C<strong>on</strong>ference</str<strong>on</strong>g> <strong>on</strong> Innovati<strong>on</strong>s <strong>in</strong> <strong>Travel</strong> Demand Model<strong>in</strong>g, Tempe, Ariz<strong>on</strong>a<br />
May 10, 2010
Outl<strong>in</strong>e<br />
•------<br />
•------<br />
•------<br />
Introducti<strong>on</strong><br />
CycleTracks for iPh<strong>on</strong>e & Android<br />
Data process<strong>in</strong>g & participants<br />
Choice set generati<strong>on</strong><br />
Model estimati<strong>on</strong> & validati<strong>on</strong><br />
Trip assignment & next steps<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 2
Why model bicycle route choice?<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 3
CycleTracks for<br />
iPh<strong>on</strong>e & Android<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
C<strong>on</strong>venti<strong>on</strong>al GPS survey<br />
… …<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 5
CycleTracks smartph<strong>on</strong>e app<br />
e-mail<br />
Trip<br />
1<br />
Trip<br />
2<br />
Trip<br />
n<br />
… … ……………<br />
…<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 6
OK, it’s real pretty;<br />
Now, how do we get users?<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 13
Data Process<strong>in</strong>g &<br />
Participants<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
GPS post-process<strong>in</strong>g (Schüssler & Axhausen 2009)<br />
5,178 traces<br />
497 users<br />
h<br />
Gaussian<br />
smooth<strong>in</strong>g<br />
Activity & mode<br />
detecti<strong>on</strong><br />
Map<br />
match<strong>in</strong>g<br />
3,034 bike<br />
stages<br />
366 users<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 18
Participants<br />
CycleTracks BATS z-stat p-value<br />
(N=366) (N=153)<br />
Age<br />
Mean 34 33 1.1 0.31<br />
Gender<br />
Female 21% 36% –3.5 0.00<br />
Cycl<strong>in</strong>g Frequency<br />
Daily 60% N/A<br />
Several times per week 34%<br />
Several times per m<strong>on</strong>th 7%<br />
Less than <strong>on</strong>ce a m<strong>on</strong>th 0%<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 19
Choice Set Generati<strong>on</strong><br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
Exist<strong>in</strong>g methods<br />
Only shortest path searches work <strong>in</strong> large networks<br />
L<strong>in</strong>k elim<strong>in</strong>ati<strong>on</strong><br />
(k shortest paths)<br />
Stochastic<br />
path search<br />
Labeled<br />
paths<br />
Doubly stochastic<br />
(Bovy & Fiorenzo-Catalano 2007)<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 21
<str<strong>on</strong>g>The</str<strong>on</strong>g> doubly stochastic method<br />
Attribute j<br />
Attribute i<br />
L<strong>in</strong>k-additive<br />
path attribute<br />
vector<br />
?<br />
1. Select random<br />
coefficient vector<br />
2. Calculate<br />
generalized cost<br />
for each l<strong>in</strong>k<br />
3. Randomize<br />
l<strong>in</strong>k costs<br />
4. F<strong>in</strong>d shortest<br />
path<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 22
Extract unbiased priors from the network<br />
Evaluati<strong>on</strong> attribute<br />
Attribute Space<br />
Reference attribute (β 1=1)<br />
Evaluati<strong>on</strong> attribute<br />
Parameter Space<br />
Value of evaluati<strong>on</strong><br />
attribute <strong>in</strong><br />
shortest path<br />
Start Coefficient with B<strong>in</strong>ary large of evaluati<strong>on</strong> boundary search attribute <strong>in</strong>terval<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 23
Results of prior extracti<strong>on</strong><br />
Significant <strong>in</strong> estimati<strong>on</strong><br />
Reta<strong>in</strong>ed for degree of freedom<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 24
96 L<strong>in</strong>k Elim<strong>in</strong>ati<strong>on</strong> Routes<br />
96 Doubly Stochastic Routes
Better
Better
Benchmark<strong>in</strong>g<br />
Method<br />
Doubly<br />
Stochastic<br />
L<strong>in</strong>k<br />
Elim<strong>in</strong>ati<strong>on</strong><br />
No. parameters 32 --<br />
No. l<strong>in</strong>k randomizati<strong>on</strong>s 3 --<br />
No. unique routes 76 (avg.) 96<br />
Search algorithm Dijkstra Euclidean A*<br />
Comput<strong>in</strong>g time 3 h 46 m 8 h 06 m<br />
2,678 observati<strong>on</strong>s, 4 CPUs, coded <strong>in</strong> Pyth<strong>on</strong><br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 28
Model Estimati<strong>on</strong> &<br />
Validati<strong>on</strong><br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
Estimati<strong>on</strong> results<br />
Attribute Coef. SE t-stat. p-val.<br />
Length (mi) --1.05 0.09 --11.80 0.00<br />
Turns per mile --0.21 0.02 --12.15 0.00<br />
Prop. wr<strong>on</strong>g way --13.30 0.67 --19.87 0.00<br />
Prop. bike paths 1.89 0.31 6.17 0.00<br />
Prop. bike lanes 2.15 0.12 17.69 0.00<br />
Cycl<strong>in</strong>g freq. < several per wk. 1.85 0.04 44.94 0.00<br />
Prop. bike routes 0.35 0.11 3.14 0.00<br />
Avg. up-slope (ft/100ft) --0.50 0.08 --6.35 0.00<br />
Female --0.96 0.22 --4.34 0.00<br />
Commute --0.90 0.11 --8.21 0.00<br />
Log(path size) 1.07 0.04 26.38 0.00<br />
2,678 weighted observati<strong>on</strong>s, ρ 2 = 0.28<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 30
Average marg<strong>in</strong>al rates of substituti<strong>on</strong><br />
MRS of Length <strong>on</strong> street for Value Units<br />
Length <strong>on</strong> bike paths 0.57 n<strong>on</strong>e<br />
Length <strong>on</strong> bike lanes 0.49 n<strong>on</strong>e<br />
Length <strong>on</strong> bike routes 0.92 n<strong>on</strong>e<br />
Length wr<strong>on</strong>g way 4.02 n<strong>on</strong>e<br />
Turns 0.10 mi/turn<br />
Total rise 1.12 mi/100ft<br />
User benefit of bike lanes: $0.98 per mile per trip<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 31
Better<br />
15% exact<br />
predicti<strong>on</strong>s
Better
Example<br />
Calibrated choice set<br />
Best generated<br />
Overlap: 85%<br />
Dissimilarity: 0.23<br />
PS-normed prob: 0.022<br />
Predicted<br />
Overlap: 5%<br />
Dissimilarity: 0.83<br />
PS-normed prob: 0.025
Trip Assignment &<br />
Next Steps<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
AM Assignment<br />
Comput<strong>in</strong>g Time:<br />
12 h 35 m<br />
Bicycles per hour<br />
0 180<br />
20 360
Next steps for SF-CHAMP<br />
Populati<strong>on</strong><br />
Synthesizer<br />
Vehicle Availability<br />
Bicycle Availability<br />
Initial Road<br />
Assignnment<br />
Roadway & Transit<br />
Skimm<strong>in</strong>g<br />
Work Loc. &<br />
Dest<strong>in</strong>ati<strong>on</strong> Choice<br />
Full Day Tour<br />
Generati<strong>on</strong><br />
Tour & Trip<br />
Mode Choice<br />
F<strong>in</strong>al Road &<br />
Transit Assignment<br />
Bike Route Choice<br />
Set Generati<strong>on</strong><br />
Logsums<br />
Logsums<br />
F<strong>in</strong>al Bicycle<br />
Assignnment<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 37
Acknowledgements<br />
Nad<strong>in</strong>e Schüssler Kay Axhausen<br />
C<strong>on</strong>tact<br />
Jeff Hood, UC Berkeley: jeffnhood@gmail.com<br />
Billy Charlt<strong>on</strong>, SFCTA: billy.charlt<strong>on</strong>@sfcta.org<br />
Matt Paul, MoPimp Producti<strong>on</strong>s: www.mopimp.com<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY
Recap<br />
•------<br />
•------<br />
•------<br />
Introducti<strong>on</strong><br />
CycleTracks for iPh<strong>on</strong>e & Android<br />
Data process<strong>in</strong>g & participants<br />
Choice set generati<strong>on</strong><br />
Model estimati<strong>on</strong> & validati<strong>on</strong><br />
Trip assignment & next steps<br />
SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY 39
Change language