The Evolution of Transport...

31
Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution of Transport Planning

Transcript of The Evolution of Transport...

Page 1: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

The Evolution of Transport Planning

Page 2: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Michael FlorianCalin D. Morosan

On Proportionality and Uniqueness inEquilibrium Assignment

Page 3: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Several bush-based algorithms for computingequilibrium assignments can obtain very finelyconverged flows Bar Gera (2002) Dial (2006) Gentile (2012)

All require post processing to obtain unique pathflows by maximizing the entropy of path flowsTAPAS method (Bar-Gera, 2010) exhibitsproportionality and hence uniqueness of path flows

Background

Page 4: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Bush-based methods are not particularly efficientfor solving multi-class network equilibrium models usually require cycling over the classes cannot be efficiently parallelized

The motivation of this investigation is the need fora more efficient multi-class traffic assignment thatcan obtain unique path and class flows

Motivation

Page 5: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

The bi-conjugate variant of the linearapproximation method solves multi-class networkequilibrium models efficientlyThe path and class flows exhibit proportionalityand hence uniqueness as the relative gapdecreasesThe linear approximation method is not as efficientbut also shares the near uniqueness properties

Findings

Page 6: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Two classes of traffic: 1->8 100 trips; 2->8 60 tripsTotal link flow indicated on the links

Path Flows and Proportionality

1

2

3

5

6

84 7

100

60

160

40

120

40

120

160

Page 7: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

These path flows are proportional25/75=15/45=40/120=1/3

Path Flows and Proportionality

1

2

3

5

6

84 7

100

60

100;60

25;15

75;45

25;15

75;45

100;60

Page 8: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

These path flows are not proportional40/60 is not equal to 0/60

Path Flows and Proportionality

1

2

3

5

6

84 7

100

60

100;60

40;0

60;60

40;0

60;60

100;60

Page 9: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

A two-class network equilibrium problem on theChicago test networkThree sets of origin-destination matrices cars and trucks three different levels of congestion

Flows obtained with the TAPAS algorithm run to arelative gap of 10-12

Available Data for Comparison

Page 10: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

1,790 zones11,192 nodes39,018 linkstwo classes cars and trucks

truck prohibition on563 links

The Chicago Test Network

Page 11: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

1 A multi-threaded implementation of the classicallinear approximation method, as implemented inEmme 4.1 Standard traffic assignment

2 A multi-threaded, conjugate direction linearapproximation method* as implemented in Emme4.1 SOLA traffic assignment The convergence of the second-order method is one

order of magnitude better with reasonable computationtimes for relative gaps of 10-5 and 10-6

Algorithms Used in the Computations

*As described in Mitradjieva, M. and Lindberg, P.O. (2013)

Page 12: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

TAPAS (10-12 relativegap) and SOLA (10-6

relative gap) havenearly identical carflowsSOLA trafficassignment runs in12.7 minutes on a 16-core, 2.9 GHz Xeonprocessor (32 threads)

Car Flows Comparison

Page 13: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

TAPAS (10-12 relativegap) and SOLA (10-6

relative gap) havenearly identical truckflowsSOLA trafficassignment runs in12.7 minutes on a 16-core, 2.9 GHz Xeonprocessor (32 threads)

Truck Flows Comparison

Page 14: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Paired Alternative SegmentsSOLA - Relative Gap 10-6

200

233

Ratio of flow onpaired alternativesegments is~200/233 = 0.8584

Page 15: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

O-D pairs contributingflow to each leg of thepair of alternativesegmentsO-D demand areobtained with a ‘select-link’ analysisSlope is~200/233 =.8584

Paired Alternative SegmentsSOLA Car flows

Page 16: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

O-D pairs contributingflow to each leg of thepair of alternativesegmentsO-D demand areobtained with a ‘select-link’ analysisSlope is~200/233 =.8584

Paired Alternative SegmentsSOLA Truck flows

Page 17: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

A comparison of the flow differences between SOLA flows at different levels of convergence TAPAS flows at 10-12 relative gap

Comparison using a log-log scale

Page 18: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Car Flows Comparison: SOLA vs TAPASSOLA 10-3 Relative Gap

Page 19: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Car Flows Comparison: SOLA vs TAPASSOLA 10-5 Relative Gap

Page 20: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Car Flows Comparison: SOLA vs TAPASSOLA 10-7 Relative Gap

RMSE=0.89

Page 21: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Truck Flows Comparison: SOLA vs TAPASSOLA 10-3 Relative Gap

Page 22: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Truck Flows Comparison: SOLA vs TAPASSOLA 10-5 Relative Gap

Page 23: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Truck Flows Comparison: SOLA vs TAPASSOLA 10-7 Relative Gap

Page 24: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

TAPAS (10-12 relative gap)and SOLA (10-6 relative gap)have nearly identical carflowsSOLA traffic assignmentruns in 17.7 minuteson a 16-core, 2.9 GHz Xeonprocessor (32 threads)

Car Flows ComparisonMore congestion

Page 25: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

TAPAS (10-12 relative gap)and SOLA (10-6 relative gap)have nearly identical truckflowsSOLA traffic assignmentruns in 17.7 minuteson a 16-core, 2.9 GHz Xeonprocessor (32 threads)

Truck Flows ComparisonMore congestion

Page 26: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

TAPAS (10-12 relative gap)and SOLA (10-6 relative gap)have nearly identical carflowsSOLA traffic assignmentruns in 32 minuteson a 16-core, 2.9 GHz Xeonprocessor (32 threads)

Car Flows ComparisonEven more congestion

Page 27: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

TAPAS (10-12 relative gap)and SOLA (10-6 relative gap)have nearly identical truckflowsSOLA traffic assignmentruns in 32 minuteson a 16-core, 2.9 GHz Xeonprocessor (32 threads)

Truck Flows ComparisonEven more congestion

Page 28: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Interesting to see how the linear approximationmethod performsStandard traffic assignment was run for 15,000iterations to achieve a relative gap of 2.3*10 -6 (notrecommended for use in practice…)3.5 hours with the linear approximation method ona 16-core, 2.9 GHz Xeon processor (32 threads)

Back to Linear Approximation

Page 29: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Car Flows ComparisonTAPAS vs Linear Approximation

RMSE=1.35

Page 30: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Truck Flows ComparisonTAPAS vs Linear Approximation

Page 31: The Evolution of Transport Planningonlinepubs.trb.org/onlinepubs/conferences/2014/ITM/Presentations/... · Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014 The Evolution

Innovations in Travel Modeling, Baltimore, MD, April 27-30, 2014

Results obtained with the linear approximationmethod and the second-order method (SOLA)exhibit near class uniqueness and proportionalityThe SOLA traffic assignment, a multi-threaded bi-conjugate variant of the linear approximationmethod, provides an attractive and computationallyefficient method for solving multi-classassignments to fine convergence

Conclusions