Professional Documents
Culture Documents
12/13/2018 1
Outline
• Introduction
• Methods in Literature
• Research Goals
• Methodology
• Location Optimization of Dedicated Bus Lanes
• Location Optimization of Transit Signal Priority
• Future work
12/13/2018 2
Introduction
12/13/2018 3
Introduction
12/13/2018 4
Introduction
12/13/2018 5
Literature Review – Transit Priority Strategies
12/13/2018 [1] Arasan and Vedagiri, 2010 [2] Gan et al., 2003 [3] Christofa and Skabardonis, 2011 6
[4] Ahn and Rakha, 2006 [5] Zyryanov and Mironchuk, 2011 [6] Zhu, 2010
Literature Review – Location Selection at Network
Level
12/13/2018 7
[1] Mesbah et al. 2010 [2] Saeednia et al., 2015 [3] Yu et al. 2015
Literature Review – Bilevel Optimization
• Lower level aims to determine the total travel time on the network given a
certain transit priority location setting
• All studies use a similar lower level algorithm
1. Travel time estimation using BPR or Akcelik function
2. Traffic assignment with a user equilibrium model based on travel times found in
step 1
3. Modal split using travel times found in step 1
4. Repeat step 1-3 until modal split values converge
• This approach has some problems
• Cost functions used for travel time estimation (BPR or Akcelik) in user equilibrium
model are not valid for saturated traffic
• It is not possible to model dynamic signal delay
• Impact of possible queue spillovers cannot be accounted with this model
12/13/2018 8
Literature Review – Bilevel Optimization
12/13/2018 10
Methodology - Overview
Upper Level
Select transit priority strategy locations
Lower Level
Determine initial user equilibrium
12/13/2018
Evaluate transit priority strategy locations 11
Lower Level – Algorithm Inputs
12/13/2018 12
Lower Level – Initial User Equilibrium
12/13/2018 13
Lower Level – Link Transmission Model
12/13/2018 14
Lower Level – Estimate Car Flows
• The link transmission model has a sending and receiving flow function
for each node
• The sending flow: Maximum number of vehicles can leave a link
• Link capacity
• Upstream traffic conditions
• The receiving flow: Maximum number of vehicles can enter a link
• Link capacity
• Jam density
• At the intersections signals are modeled by reducing the sending flow
to 0 for the duration of red signal
12/13/2018 15
Lower Level – Estimate Car Travel Times
Cumulative number
Downstream
𝑡1 : entry time
downstream and upstream cumulative vehicle 𝑡2 : exit time
diagrams of links.
Car travel time
4𝑚𝑖𝑛 𝑖 2𝑚𝑖𝑛
4+5+2+1=12min
𝑇𝑇𝑖𝑗 = 10𝑚𝑖𝑛
2+1+4+3=10min
12/13/2018 16
Lower Level – Estimate Bus Travel Times
Cumulative number
Downstream
assumed that bus has left the link 𝑛 : initial bus number 0
• Bus travel times between each OD pair is estimated with 𝑛0 𝑛1 : updated bus number
𝑡0 : free-flow travel time
𝑛1
the same assumption used for car travel times Travel time
𝑡 𝑡+𝑡0 Time
12/13/2018 17
Lower Level – Route and Mode Choice
12/13/2018 18
Lower Level – Calculate Total Travel Time
• At the end of LTM run total travel time of network users is calculated
for upper level optimization
• The last 15 minutes of the simulation is assumed to be representative of
converged mode choice and route choice
• For the last 15 minutes:
• The car total travel time is calculated by considering the cumulative entry and
exit diagrams of the network determined at the origins and destinations
• The bus travel time is calculated from total bus demand between each OD
pair and in-vehicle bus travel time between each OD pair
12/13/2018 19
Upper Level – Genetic Algorithm
12/13/2018 20
Upper Level – Select Parent and Create Offspring
• Parent selection and creation of offspring are the main steps that
guide the algorithm to a better solution.
• For parent selection a tournament selection procedure is used to
create a balanced selection probability among parent candidates.
• To create offspring chromosomes, reduced surrogate crossover
method is used to decrease the probability of producing offspring
identical to the one of the parents
12/13/2018 21
Upper Level – Generate Next Generation
12/13/2018 22
Location Optimization of Dedicated Bus Lanes
12/13/2018 23
Tested Demand Patterns
12/13/2018 24
List of Tested Scenarios and Used Methods
Genetic algorithm
Genetic Enumeration with
Demand Pattern and the Network with symmetry
algorithm symmetry constraint
constraint
1stst network – Saturated uniform demand √ √
1stst network – Saturated evening peak demand √
1stst network – Saturated morning peak demand √
2nd network – Saturated uniform demand √ √
2nd network – Saturated evening peak demand √
2nd network – Saturated morning peak demand √
1st network – Undersaturated uniform demand √
1st network – Undersaturated Saturated evening peak
√
demand
1st network – Undersaturated Saturated morning peak
√
demand
2nd network – Undersaturated Saturated uniform demand √
2nd network – Undersaturated Saturated evening peak
√
demand
2nd network – Undersaturated Saturated morning peak
√
demand
12/13/2018 25
Genetic Algorithm – Uniform Saturated Demand
Total travel Total bus travel Total car travel Mode share of Mode share
time (hr) time (hr) time (hr) bus (%) of car (%)
No bus lane 664.4
664.4 277.4 387.0 42.3%
42.3% 57.7%
57.7%
Optimum
configuration 656.2
656.2 260.4 395.8 42.4%
42.4% 57.6%
57.6%
Full buslane 920.9
920.9 234.0 687.0 42.8%
42.8% 57.2%
57.2%
12/13/2018 26
Enumeration of symmetric configurations
4
9 3
5 4
5 10 11 12
4 5 6
8 2
4 5 5
7 1
4
Total travel Total bus Total car travel Mode share of Mode share of Total travel Total bus Total car travel Mode share of Mode share of
time (hr) travel time (hr) time (hr) bus (%) car (%) time (hr) travel time (hr) time (hr) bus (%) car (%)
No bus lane 664.4 277.4 387.0 42.3% 57.7% No bus lane 664.4 277.4 387.0 42.3% 57.7%
Optimum Optimum
configuration 647.3
647.3 257.7 389.7 42.5% 57.5% configuration 656.2
656.2 260.4 395.8 42.4% 57.6%
Full buslane 920.5 233.6 687.0 42.8% 57.2% Full buslane 920.9 234.0 687.0 42.8% 57.2%
12/13/2018 Result of GA 28
Enumeration – Morning Peak Demand
Total travel Total bus Total car travel Mode share of Mode share of Total travel Total bus Total car travel Mode share of Mode share of
time (hr) travel time (hr) time (hr) bus (%) car (%) time (hr) travel time (hr) time (hr) bus (%) car (%)
No bus lane 665.5
665.5 278.0 387.5 42.9% 57.1% No bus lane 664.4
664.4 277.4 387.0 42.3% 57.7%
Optimum Optimum
configuration 652.9
652.9 262.7 390.2 43.1% 56.9% configuration 647.3
647.3 257.7 389.7 42.5% 57.5%
Full buslane 999.6
999.6 228.9 770.7 43.7% 56.3% Full buslane 920.5
920.5 233.6 687.0 42.8% 57.2%
Total travel Total bus Total car travel Mode share of Mode share of Total travel Total bus Total car travel Mode share of Mode share of
time (hr) travel time (hr) time (hr) bus (%) car (%) time (hr) travel time (hr) time (hr) bus (%) car (%)
No bus lane 664.4
664.4 277.4 387.0 42.3% 57.7% No bus lane 664.4
664.4 277.4 387.0 42.3% 57.7%
Optimum Optimum
configuration 647.3
647.3 257.7 389.7 42.5% 57.5% configuration 647.3
647.3 257.7 389.7 42.5% 57.5%
Full buslane 920.5
920.5 233.6 687.0 42.8% 57.2% Full buslane 920.5
920.5 233.6 687.0 42.8% 57.2%
12/13/2018 31
𝑆=1 𝑆=2 𝑆=3 𝑆=4
𝑆=9 𝑆 = 10 𝑆 = 11 𝑆 = 12
12/13/2018 32
Location Optimization of Transit Signal Priority
12/13/2018 33
TSP activation
12/13/2018 34
Evaluation of All TSP case
12/13/2018 35
Results of location optimization
# of intersections Total
Total bus
bus Total Car Total # of times
equipped with travel
travel travel travel TSP
Scenario TSP time
time (hr)
(hr) time (hr) time (hr) activated
a (Baseline) 0 568
568 815 1383 0
b 4 504
504 813 1317 31
c 4 529
529 813 1342 15
d 4 554
554 814 1368 27
e 4 552
552 814 1366 27
f 8 442
442 808 1250 73
g 8 543
543 814 1356 53
h 8 498
498 809 1307 70
i 8 496
496 809 1305 74
j 8 503
503 816 1319 31
k 8 499
499 815 1314 32
l 12 477
477 808 1285 116
m 12 505
505 815 1319 46
n 12 424
424 807 1231 109
o 12 424
424 809 1233 110
p 16 407
407 807 1214 140
12/13/2018 36
Future Work
12/13/2018 37
Optimization for Transit Signal Priority
12/13/2018 38
Optimization for Intermittent Bus Lanes
• The current model does not allow dynamic control of the status of
the lane
• The links with intermittent bus lanes must be modeled with two links
• Intermittent bus lane can be activated by using the tracking algorithm used
for TSP activation
12/13/2018 39
Alternative Algorithms for Upper Level
Optimization
12/13/2018 40
Finding Optimum Sequence of Implementation
12/13/2018 41
Research Timeline
12/13/2018 42