Upload
meena
View
36
Download
0
Embed Size (px)
DESCRIPTION
Parsimony in Advanced Transportation Forecasting . We are to admit no more causes than such as are both true and sufficient to explain their appearances. TRB Planning Applications Conference May 8 - 12, 2011 Reno, NV Stephen Lawe Resource Systems Group Maren Outwater - PowerPoint PPT Presentation
Citation preview
Parsimony in Advanced Transportation Forecasting
We are to admit no more causes than such as are both true and sufficient to explain their appearances
TRB PlanningApplications ConferenceMay 8 - 12, 2011Reno, NV
Stephen LaweResource Systems Group
Maren OutwaterResource Systems Group
Joe CastiglioneResource Systems Group
2
Rational for Advanced Models
Traffic shifts by time-of-day Peak spreading, peak shifting
Tolling and Pricing Impacts Tolling, congestion pricing
Travel time reliability effects Behavioral response to reliability
Operations impacts Signals and coordination & ITS
Travel Demand Management Flexible work schedules, Work / shop at home
The policy questions facing the industry have changed:
3
Definition: Advanced Models
For the purpose of this presentation “Advanced Travel Modeling” will be defined as:
Spatially and temporally disaggregate model systems that represent behaviorally realistic individual or agent-based decision making
4
Definition: Parsimony
Parsimony:
1. Unusual or excessive frugality; extreme economy or stinginess
2. Economy in the use of means to an end
5
Parsimony in Scientific Thinking
Aristotle: Lex Parsimoniae – used in Posterior Analytics as an Ontological Principle
“We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances.”
Isaac Newton
“That is done in vain by many means which may equal well be done with fewer”
Nash (not from A Beautiful Mind)
6
Principle of Parsimony - Lex Parsimoniae
A methodological principle dictating a bias towards simplicity in theory construction, where the parameters of simplicity vary from kinds of entity to the number of presupposed axioms to characteristics of curves drawn between data points.
William Ockham(Ockam’s Razor)
7
Principle of Parsimony applied to Travel Modeling
This does not mean: the simplest explanation is most likely the correct
one
Instead it means: We should tend towards simpler theories until we can
trade some simplicity for increased explanatory power
Application to Transportation Modeling1. Model structural detail should be consistent with
our understanding of behavior2. “Simpler Theories” are often those which are
behaviorally descriptive and explainable- Conventional models (eg. 4-step) do not necessarily
represent “Simpler Theories”
8
Linked Model System
Develop a linked model in Jacksonville, FL and Burlington, VT
DaySim: Provides detailed estimates of travel demand TRANSIMS: Provides detailed estimates of network
performance MOVES: Provides detailed estimates of air quality
LOS Skims
Demand FileTRANSIMS STUDIO
Iteration/ConvergenceFile Manager
DaySim Exogenous Trips
TRANSIMS
MOVESMOEs / Indicators
9
Spatial & Temporal Resolution
SPATIAL RESOLUTION Fine Intermediate Coarse
Basic spatial unit Parcels Activity locations Zones
Model systems DaySim TRANSIMS DaySim & TRANSIMS
Number in Jacksonville 620,000 30,000-220,000 1,300
TEMPORAL RESOLUTION Fine Intermediate Coarse
Basic temporal unit Seconds Minutes Hours
Model systems TRANSIMS Microsimulator
DaySim <-> TRANSIMS DaySim & TRANSIMS
Interval in Jacksonville 1 second 1-15 minutes 1-24 hours
Multiple spatial and temporal resolutions used in the model system
10
Spatial Resolution
Parcels are core spatial choice unit in DaySim
Activity locations are basic spatial unit in network assignment
Parcel level land use and socioeconomic data requires significant data review and cleaning
Traffic controls and other spatially detailed roadway network configurations require review and adjustment
11
Network Resolution
Planning network
Networks in Jacksonville at 3 different spatial resolutions
Finegrained network All streets network
9,865 links 29,284 activity locations
15,933 links 51,292 activity locations
96,362 links 216,350 activity
locations
12
All Streets Jacksonville Network Resolutions
13
Temporal Resolution
Current skim resolution: 22 time periods, TAZ-level Future skim resolution: 48 time periods, “activity-level”
resolution
1 overnight skim 9 hourly midday & shoulder skims 12 30-min peak period skims
EV PM EVAM MD
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
% o
f Reg
iona
l Tra
vel
14
Local vs. Regional: Tour & Trip Time-of-Day
Local Test Regional Test
Trip
To
ur
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Tour Departure Time (1/2 hour): Local Test
V4V22
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Tour Departure Time (1/2 hour): Regional Test
V4V22
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Trip Departure Time (1/2 hour): Local Test
V4V22
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Trip Departure Time (1/2 hour): Regional Test
V4V22
15
Convergence
Condition where experienced impedances and costs produced by the model are approximately equal to expected impedances and costs input to the model and overall demand flows are stable
Necessary for model system integrity and policy sensitivity
Applies to both network assignment convergence and overall model system convergence
16
Low vs. High Iteration: Tour & Trip Time-of-Day
Low Iteration (3x25) High Iteration (6x40)
Trip
To
ur
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Tour Departure Time (1/2 hour): Local Test
V4V22
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Trip Departure Time (1/2 hour): Local Test
V4V22
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Tour Departure Time (1/2 hour): Local Test
V4V22_HI
-0.3%
-0.2%
-0.1%
0.0%
0.1%
0.2%
0.3%
02:3
003
:30
04:3
005
:30
06:3
007
:30
08:3
009
:30
10:3
011
:30
12:3
013
:30
14:3
015
:30
16:3
017
:30
18:3
019
:30
20:3
021
:30
22:3
023
:30
00:3
001
:30
Change in All Trip Departure Time (1/2 hour): Local Test
V4V22_HI
17
Spatial TripGap Measure (TAZ-level)
V4 DaySim 3 system iterations 25 assignment iterations
below - 0.010
0.010 - 0.030
0.030 - 0.050
0.050 - 0.100
0.100 - above
18
V22 DaySim 6 system iterations 40 assignment iterations
below - 0.010
0.010 - 0.030
0.030 - 0.050
0.050 - 0.100
0.100 - above
Spatial TripGap Measure (TAZ-level)
19
Network Convergence Lessons Learned
“problem” trips undermine the purpose and integrity of the model and convergence metrics
Different methods will converge to different total system costs
20
Integrated Model System
Integrated Activity Based model with Microsimulator
21
System Convergence
System convergence definition and measures for advanced integrated models not well established
Demand convergence: changes across system iterations should be minimized
Impedance convergence: inputs should be approximately equal to outputs
Link flow/cost convergence
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
ITER_1 ITER_2 ITER_3 ITER_4
% R
MSD
Iteration
%RMSD Trip by District
RADSUABREAPUMA
22
Basic Observations of Stability
In the model system we are testing we have observed a general region-wide stability with:
– 40 microsimulator iterations (internal iterations)– 6 system iterations (system iterations)
1. It is not clear, however, whether this is sufficient for localized project-level stability.
2. Furthermore, as increased detail is added, further iterations are warranted.
3. Finally, it may well be the case that some level of detail such as operations level network detail may never converge in a microsimulator context
23
Considerations of Parsimony
Policy motivation for increased detail
Model stability & forecasting confidence
Data support and behavioral defensibility
Understandability and use in policy context
Consideration of Parsimony
24
Stephen LaweManaging DirectorResource Systems [email protected]
May 8, 2011
Thank You and Happy Mothers Day