Transcript Presentation - 15th TRB National Transportation Planning

Extending the Random-Utility-Based
Multi-Regional Input-Output Model
(RUBMRIO) to the National Level
Dr. Kara Kockelman
Dr. Avinash Unnikrishnan
Siva Karthik Kakaraparthi
The RUBMRIO Model
• Combines three distinct models:
 Input-output for production
 Logit model of input origins & modes
 Traffic assignment
• Recognizes:




18 economic sectors
3,109 U.S. counties
Nation’s top 34 Ports
Two modes: highway & rail
Industry Sectors
1 Agriculture, Forestry, Fishing & Hunting
2 Mining
3 Construction
4 Food Manufacturing
5 Chemicals Manufacturing
6 Primary Metals Manufacturing
7 Fabricated Metals Manufacturing
8 Machinery Manufacturing
9 Electronic & Electric Equipment
10 Transportation Equipment
11 Other Durable & Non-Durable Manufacturing
12 Transportation, Communications & Utilities
13 Wholesale Trade
14 Retail Trade
15 FIRE (Finance, Insurance & Real Estate)
16 Service
17 Households
18 Government
The RUBMRIO Model (2)
Inputs: Foreign Exports, Domestic Exports, Distances
Utility of purchasing
commodity m from zone i
and transporting it to zone j, k, s
Flow of m from zone j to
export zone k, and other state s
Production of m in zone i
Consumption of m in zone j
m
m
U ijm   pim  m ln[exp(0m   highway
 dij ,highway )  exp(  railway
 dij ,railway )]
m
m
U ikm   pim  m ln[exp(  0m   highway
 d ik ,highway )  exp(  railway
 d ik ,railway )]
m
m
U ism   pim  m ln[exp(0m   highway
 dis ,highway )  exp(  railway
 dis ,railway )]
Yikm  Ykm
exp(U ikm )
 exp(U ikm )
Z ism  Z sm
exp(U ism )
 exp(U ism )
No
i
i
xim   X ijm   Yikm   Z ism
j
k
s
Trade
Equilibrium?
n
C mj   ( Amn
j  xj )
n
Yes
Flow of m as an intermediate input
from zone i to zone j
Average input cost for
commodity m in zone j
X C
m
ij
m
j
exp(U ijm )
 exp(U
m
ij
 X U

X
m
ij
c
m
j
m
ij
i
m
ij
i
Price of m in zone j
)
i
p nj   ( A0mnj  c mj )
m
Equilibrium Trade Flows
Data Sets
• Imports & Exports for nation’s top 44 ports
available at USATradeOnline.gov.
• Top 34 ports, each exporting >$100
million/yr., were selected from 44 available.
• Caliper Corp. provided US highway network.
• Rail network obtained from Census 2000.
• Inter-county & county-to-port shortest
paths by road & rail computed using
TransCAD.
Other Data Sources
Data Sources
Logit Trade Coef.
Truck Trips
Work Trips
Shop Trips
• 1997 CFS
• 1997 CFS
• Vehicle Inventory & Use Survey
• 1997 IMPLAN Data
• 2000 Census Commute Data
• 1990 Household Expenditures (BEA)
• Nationwide Personal Transportation
Survey (NPTS)
• Austin Travel Survey of Households
Export Ports
List of Ports
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
11)
12)
13)
14)
15)
16)
17)
Baltimore, MD
Boston, MA
Buffalo, NY
Charleston, NC
Chicago, IL
Cleveland, OH
Columbia-Snakes, OR
Dallas-Fort Worth, TX
Detroit, MI
Duluth, MN
El Paso, TX
Great Falls, MT
Houston, TX
Laredo, TX
Los Angeles, CA
Miami, FL
Minneapolis, MN
18) Mobile, AL
19) New Orleans, LA
20) New York, NY
21) Nogales, AZ
22) Norfolk, VI
23) Ogdensburg, NY
24) Pembina, ND
25) Philadelphia, PN
26) Port Arthur, TX
27) Portland, OR
28) San Diego, CA
29) San Francisco, CA
30) Savannah, GA
31) Seattle, WA
32) St. Albans, VT
33) Tampa, FL
34) Washington, D.C.
Total Exports (Million $)
US Highway Network
Network with County Centroids
US Rail Network
Road & Rail Network Details
Highway Network
• 359,328 links & 136,930 nodes (+3,109
CCs)
• Presently streamlining network assignment
code (in C++) to run efficiently on such a
large network.
Rail Network
• 156,172 links & 119,176 nodes
• 100-mile connectivity distance imposed.
Traffic Assignment
• Vehicle Trips due to economic flows…
•
 Truck & Rail Trips, Work trips & Shop
trips
Highway network assignment based on
1-hour user equilibrium procedure:
PEAKVehij  0.05SHTrips ij  0.20WRKTrips ij  0.08PCE  TTrips ij
• Production costs (& thus input prices)
updated based on updated/congested
travel times.
Code Progress
• All Windows functions modified to generic
C++ functions.
• RUBMRIO code now runs on Windows &
Linux.
• RUBMRIO runs for Texas network (254
counties) on standard desktop.
• Texas Advanced Computing Center’s
supercomputers will be used to run
national applications (3,109 counties).
Code Progress (2)
• More efficiently coded to reduce memory
allocations (rather than defining new arrays).
• Original twelve 3-D arrays of 183M units each
 Memory alloc. = 183M × 8 (double) bytes
= 1.4 GB of space per array.
• Code streamlined to now use only three
arrays of the same size.
• 68% reduction in RAM use.
Example Results: Texas Application
Year 2010
Year 2020
Example Results (2)
Variations in Intra-County Trade
Counties close to high-producers exhibit
greatest % reductions, as they increase
trade to those counties.
Coming Work
• Memory-efficient traffic assignment code is
necessary to recognize network congestion for
trade-cost implications.
• Once assignment code is complete, trade flows
between all U.S. counties will be modeled.
• Case studies include impact of network
improvements & disruptions on trade flows, impacts
of growing export demands & shifts in commodity
types demanded, & effects of system management
(e.g., pricing).
Thank You!
Questions & Suggestions?
RUBMRIO Papers available at
www.ce.utexas.edu/prof/kockelman/home.html