Transcript Slide 1

The New Trend of Travel Demand Model
Lessons learned from the New York Best Practice Model
Kuo-Ann Chiao
Director of Technical Services
New York Metropolitan Transportation Council
Many years ago at NYMTC…..
We used a Mainframe computer
to run our travel demand model
The problem was..
When there was a problem,
Even we took apart the
computer,
Or
Dig a hole on the floor,
It was very difficult to find out
what was the problem,
And
Where was the problem..
The old fashion computer print
outs did not help us too much
NYBPM Study Area
• 20,000,000 population
• 100 population segments
• 4,000 Transportation Analysis Zones
• 4 time periods
• 6 trip purposes
• 10 motorized modes
• 4 urban types
Location Distribution
1997 Household Travel Survey
A joint project between NYMTC and NJTPA
Location-based
11,000 households
28,000 people
118,000 trips
Highway Network
Uni-directional coding & Ramps
Very large network (52,794 links in 28 county 3 state NY
metro area)
4,950 High-level facilities
26,385 Arterials
10,694 Centroid and external connectors
10,765 Other
Unidirectional / dualized coding
Conflated the network geography
GIS Street Network – TIGER (or LION) Developed in
TransCAD Software
SOV, HOV2, HOV3+, taxi, truck, other commercial
Classified by 21 Primary Link Types for capacities, initial
speeds and VDF’s
Link Attributes
Zones System–Census Tract Based
BPM zone boundaries
Transit Network
Extremely detailed transit coding based
on information from MTA and NJ Transit
Developed in TransCAD 4.0
Each route variation coded as a distinct
route:
100 NYC subway routes
900 Commuter rail routes
2,300 bus routes
73,000 transit stops.
50 ferry routes
Includes sidewalk network in Manhattan
Walk access/egress links
Park - and - Ride
Transit Network
Highlights of NYBPM
Micro-Simulation choice models
Population synthesis and intrahousehold travel interactions
Journey-based travel units modeled
Non-motorized (pre-mode choice)
Mode-Destination Choice (nested
logit)
Stop frequency and location submodel
Full multi-modal analysis / assignment
Route-Deviation Concept
Someone give me an example of how do you come to the office today.
Route-Deviation Concept
Stop
dik
k
Origin
i
dkj
Journey
Destin
dij
j
Trip
A journey reflects the real travel characteristics
It also reduces the number of trip purposes needed
Micro-Simulation
General Modeling Structure
Journey Generation
Mode & Destination
Time of Day
Assignment
Journey Generation
5-Percent Census Public Use Microdata Sample (PUMS) Files
Journey Generation
Mode & Destination
Synthetic
Population
Auto
Ownership
Time of Day
Journey
Frequency
Assignment
Seed PUMS
SocioEconomic
Targets
LUM
Accessibility
Journey Frequency Model
Intra-Household Interaction
NonWorkers
Children
Mandatory
School
School
University
School
University
Work
At Work
Discretionary Maintenance
Individual Time-Space Constraint
Workers
M
M
M
D
D
D
Mode & Destination
NM
Dest.
Mot.
Dest.
Mode
Time of Day
Stop Frequency
Stop Location
Assignment
Density
Land Use Attractors
Mode & Destination
Pre-Mode
LOS Skims
Journey Generation
Pre-mode Choice: Nested Structure
Non-motorized
Mode
Motorized
Mode
Density of
attractions
Destination
Choice
Purpose-specific
attractions
Transit/
Drive Alone
Shared Ride
Taxi
Mode & Destination Choice
Pre-Mode Choice
NonMotor
ns
e
D
ity
Destination
Motorized
Destination
Accessibility
NonMotorized
Mode
Mode
Chain
Utility
Stop Frequency
Stop-Density
Total
Activity
Control
Stop Location
Detailed
Sub-Mode
Stop Frequency by Purpose
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Work- Work- Work- School Univ
low
med high
No stops
Outbound
At Maint Discr
work
Return
Both
Stop Frequency by Mode
80%
70%
60%
50%
40%
30%
20%
10%
0%
Drive
alone
Shared
ride
No stops
Transit Commut
rail
Taxi
Outbound
Return
School
bus
Both
Other
Stop Distribution by Duration
80%
70%
60%
50%
40%
30%
20%
10%
0%
<1h
1-2 h
2-3 h
3-4 h
Activity duration, hours
4-5 h
>5h
Mode & Destination Choice
Pre-Mode Choice
NonMotor
ns
e
D
ity
Destination
Motorized
Destination
Accessibility
NonMotorized
Mode
Mode
Chain
Utility
Stop Frequency
Stop-Density
Total
Activity
Control
Stop Location
Detailed
Sub-Mode
Stop-Frequency Choice Model
Choice Alternatives
Structural Dimensions
3 - Both
2 - Return
0 - No stops
1 - Outbound
Journey Purpose
Work
School
University
Maintenance
Discretionary
At Work
Person Type
...
...
Utility Components
Income
Car Sufficiency
Household
Composition
Journey Distance
...
Worker
Child
Other Journeys
Journey Purpose
Non-Worker
Stop-Location
(Density) Log-Sum
Work
School
Mode
University
Maintenance
Discretionary
SOV, Taxi
HOV
Transit
Stop-Location Choice Model
Choice Alternatives
5 miles
Structural Dimensions
Journey Purpose
Work
School
Stop Density (Size)
At Work
Maintenance
University
Combined
Impedance
Discretionary
Journey Leg
Outbound
20%
Utility Components
Route Deviation
Return
Person Type
Worker
Child
Non-Worker
Mode
5 miles
SOV, Taxi
Stop Activity
HOV
Transit
Work
School
University
Maintenance
Discretionary
Time of Day
Journey Generation
Mode & Destination
Stage 1:
Time of Day
Stage
2:
Journey
Split
by
(Current)
Stage
3:
Legs
and
Journey
TOD
Periods
Split by
Choice
Trips and
Model
Periods
Assignment
Predetermined
Set of TOD
Distributions
Timing &
Durational
Utility
LOS Skims
Stages of Calibration
and Validation Sources
Disaggregate Calibration
Household Survey
by Purpose
Aggregate
calibration Calibration
Household Survey;
n : the act of checking or adjusting (by comparison with a standard) the
Of
Destination
Choice
PUMS
accuracy
of an estimated
coefficients.
Aggregate
validation Calibration
Household Survey;
n : the act of finding or testing the truth of something
PUMS
Of Mode Shares
Highway and Transit
Assignment
Traffic Counts;
Screenline Database;
MATRIX; HPMS
Fractional Probability
Mode 1 (0.05)
Destination 1 (0.15)
Mode 2 (0.03)
Mode 3 (0.07)
Mode 1 (0.15)
Tour
Destination 2 (0.75)
Mode 2 (0.25)
Mode 3 (0.35)
Mode 1 (0.05)
Destination 3 (0.10)
Mode 2 (0.02)
Mode 3 (0.03)
Micro-Simulation
X
Destination 1 (0.15)
X
Mode 2 (0.25)
X
Mode 1 (0.15)
Tour
Destination 2
Mode 3
X
Destination 3 (0.10)
Aspects of Micro-Simulation
for NYBPM Processing
Nearly 9 million households in base year
Journey productions file over 500 Meg
Mode destination choice stops model
processes over 25 million paired journeys by 8
trip purposes
Output files over 300 Meg
6 highway classes and 4 transit trip tables for
each of 4 time periods
Combined file size about 2.5 Gig
Hardware: 4 GB RAM / Dual Processor / 1.5
Ghz / 120+ GB Hard Drive
Dimension of Choice Probability in NYBPM
Core Destination & Mode Choice Probability
Destination
Origin
D=1
O=1
D = 2 ...
Mode = 1
D = 4000
Mode = 2 ...
Mode = 10
O = 2 ...
O = 4000
Stratum = 1,2 ... 100
Conventional Fractional
Probability Array
Micro-Simulation
20,000,000 Stratified Individuals
(0=4000)*(D=4000)*
(M=10)*(S=100) =
23,000,000 Origin-Based Journeys
16,000,000,000
23,000,000 OD-Based Mode Choices
23,000,000 Destination Choices
Processing Time For BPM Model Run
STEP
BPM PROCEDURE
HOURS
1
CREATE NEW SCENARIO
10 min.
2
RUN HIGHWAY NETWORK BUILDER
15 min.
5 min.
3
NETPREP
.20 min.
4 min.
4
HIGHWAY PRESKIMS
12 hrs.
5 hrs 30 min.
5
TRANSIT NETWORK DATABASE & SKIMS
48 hrs.
12 hrs 6 min.
6
ACCESSIBILITY INDICIES
2 hrs.
26 min.
7
HOUSEHOLD AUTO JOURNEY (HAJ)
1 hrs.
4 min.
8
MODE DESTINATION STOPS CHOICE (MDSC)
18 hrs.
9
TRUCKS/COMMERCIAL VEHICLES MODEL
2 hrs.
10
EXTERNAL MODEL
5 min.
11
PRE-ASSIGNMENT PROCESSING/TIME OF DAY (PAP)
1 hrs.
12
HIGHWAY ASSIGNMENT
16 hrs.
13
TRANSIT ASSIGNMENT
72 hrs. 42 hrs 40 min.
TOTAL
9 hrs 45 min.
6 hrs 43 min.
173 hrs
(>7 days)
current improvements
78 hrs
(3 days)
Status of On-Going Improvements
Speed up the running time
Software Engineering
Memory Handling
allocated the memory only once, using a flag to determine if the memory had
already been allocated
memory could be allocated in one block
Input/Output
Remove messages (one per 33 million lines in the HAJ trip file) to the screen,
reduced processing time from 22 minutes to 20 seconds
Parameter Passing
Passing information of a pointer to a structure rather than an entire structure
(e.g., the memory used to call about 260,000 times of one function with 92 bytes
could be reduced significantly by passing a pointer to the structure that only
requires 4 bytes)
In-lining Function Calls
Very short functions that are called frequently can cause bottlenecks (function
consists of just a few lines (e.g., Calling a function, which was being called
between 300,000 to 600,000 times, was taking up 10% of the total program time.
In-lining the function reduced it to 0.3% of the total program time)
Additional optimization
Hardware optimization
BPM Structure –“GUI” for User Documentation
Applications of BPM at NYMTC
Conformity Analysis
Regional Transportation Plan
Congestion Management Systems
Testing Scenarios for emission reduction
strategies
Request for Data Manipulation and Runs from
other agencies
Applications of BPM .. Projects
Tappan Zee Bridge
Gowanus Expressway
Bronx Arterial Needs
Bruckner Sheriden Expressway
Long Island East Side Study
Canal Area Transportation Study
Lower Manhattan Development Corporation
Southern Brooklyn Transportation Study
Regional Freight Plan Study
Hackensack Meadowland Development Corp.
Model Update
Study of Post 9/11 Travel Pattern Changes
New Set of Socioeconomic and
Demographic Forecasts
Collection of 2002 traffic and transit data
Updated 2002 base year Model
Model Improvements
Better Highway -Transit Connection
Improve transit models
Integrate BPM with the Land Use Model
Web Applications
Model output analysis
Model runs
Distributed Process
Better GUI (flowchart-based, on-line help & document)
More project applications
BPM User’s Group Support & Meetings
Advisory Committee
Patrick T. Decorla-Souza, Federal Highway Administration
Frederick W. Ducca, Federal Highway Administration
Ron Jensen-Fisher, Federal Transit Administration
Elaine Murakami, Federal Highway Administration
Bruce Spear, Federal Highway Administration
John Thomas, Environmental Protection Agency
T. Keith Lawton, Metro, Portland, Oregon
Charles Purvis, Metropolitan Transportation Commission
David Zavattero, Chicago Area Transportation Study
Arnim H. Meyburg, Cornell University
The Late Eric Paas, Duke University
Frank Spielberg, S. G. Associates
That’s it for today
Thanks 