Modelling Passengers’ Route Choice Behaviour on the London Underground: Application of Two Choice Modelling Approaches Tamas Nadudvari [email protected]Dr Ronghui Liu [email protected]Professor Stephane Hess [email protected]Nadudvari, Liu, Hess ITS Uni of Leeds UTSG 2015, City University London 06 January 2015
26
Embed
Modelling passengers’ route choice behaviour on the london underground
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Modelling Passengers’ Route Choice Behaviour on the London Underground: Application of Two
• Application of choice modelling approaches– Bayesian Modelling Framework (BMF)
– Random Utility Maximisation (RUM)
• Conclusion– Conclusion
– Further research
Nadudvari, Liu, HessITS Uni of Leeds
Introduction
UTSG 2015, City University London
06 January 2015Source: TfL
Source: newtravelco.com
Where are the passengers in the network? How can I avoid the crowd?
How can I use Smartacard datato answer these 2 questions?
Nadudvari, Liu, HessITS Uni of Leeds
Objectives
UTSG 2015, City University London
06 January 2015
Transit Assignment Model
Route Choice Model Path Generation Model
• Bayesian Modelling Framework (BMF) (Fu 2014)FU, Q., LIU, R. & HESS, S. 2014. A Bayesian Modelling Framework for Individual Passenger’s Probabilistic Route Choices: a Case Study on the London Underground. Transportation Research Board (TRB) Annual Meeting.
oWhich route passengers have chosen?oInfer route choice from Observed Journey Time (OJT)
Apply for case study network Compare results Apply in TAM
Nadudvari, Liu, HessITS Uni of Leeds
Objectives
UTSG 2015, City University London
06 January 2015
Transit Assignment Model
Route Choice Model Path Generation Model
• Random Utility Maximisation (RUM) (Raveau et al 2014)RAVEAU, S., GUO, Z., MUÑOZ, J. C. & WILSON, N. H. M. 2014. A behavioural comparison of route choice on metro networks: Time, transfers, crowding, topology and socio-demographics. Transportation Research Part A: Policy and Practice, 66, 185-195.
oWhich route passengers would choose to maximise utlity?oUtility from attributes (time, transfer, crowding, topology, socio-demographics)
Apply for case study network Compare results Apply in TAM
Nadudvari, Liu, HessITS Uni of Leeds
Case Study Network
UTSG 2015, City University London
06 January 2015
Clapham Junction
EastPudney
LU
SC
Tooting
London Underground (LU)
Northern line•Via Bank•Via Charing Cross
Services•A: From Morden, via Bank (2-4 min)•B: From Morden, via CX (10-15 min, peak)•C: From Kennington via CX (3 min)
Origin zone: Morden – OvalDestination zone : Waterloo – Goodge Street
Route 1: Direct (Service B)Route 2: Change at Kennigton (Service A+C)Morden
Kennington
Bank
CharingCross
Euston
C
C
C
B
B
B
B
A
A
AA
Morden
South Wimbledon
Colliers Wood
Tooting Broadway
Tooting Bec
Clapham Common
Balham
Clapham South
Clapham North
Stockwell
Oval
Waterloo
Tottenham Court Road
Embankment
Charing Cross
Leicester Square
Goodge Street
Kennington
B
B
B
A
C
C
Nadudvari, Liu, HessITS Uni of Leeds
Initial Data
UTSG 2015, City University London
06 January 2015
• 5% Individual Oyster data, 4 week (06/02-05/03/2011) → 2676 transactions of 153 regular commuters (min 15 days)Case study network (Northern line), weekday, AM peak
• Timetable data https://www.whatdotheyknow.com/request/london_underground_timetables
• Access Egress Interchange (AEI) data (06/02-05/03/2011)
• Station layout, Direct Enquires (DE) http://www.directenquiries.com/londonunderground.aspx
• Rolling Origin and Destination Survey (RODS) (1998-2010) →6330 respondents for case study network
FU, Q., LIU, R. & HESS, S. 2014. A Bayesian Modelling Framework for Individual Passenger’s Probabilistic Route Choices: a Case Study on the London Underground. Transportation Research Board (TRB) Annual Meeting.
Nadudvari, Liu, HessITS Uni of Leeds
Oyster Journey Time (OJT) frequencies
No data
Few dataUnable to fit distribution
Zonal Journey Time (ZJT) frequencies
Bigger dataset better to fit distribution
Bayesian Modelling Framework (BMF)
UTSG 2015, City University London
06 January 2015
Zonal Journey Time (ZJT)
Tent Tex
tent-CO tex-CD
ZJT
ZJT: Journey time from zone centroid to zone centroid.Tent /Tex: Entry/Exit time, Oyster data.tent-CO / tex-CD: In veh. time bween entry/exit station and centroid, timetable∆tacc/∆tegr: Correction due to diff. acc/egr times at stations, AEI/DE
(Tex+tex-CD +∆tegr)-(Tent )+tent-CO+∆tacc
Station
Zone centroid
ZJT=
Nadudvari, Liu, HessITS Uni of Leeds
Bayesian Modelling Framework (BMF)
UTSG 2015, City University London
06 January 2015
ZJT frequencies
ZJT [min]
Oys
ter
tran
sact
ion
s [#
]
Supposing two routes
Setting a Gaussian mixture distribution of two components (default case)
Calculating parameters
Mean SD Probability
[min] [min] [%]
Route 1 27.05 9.33 86%
Route 2 32.87 30.97 14%
Comparing with Scheduled Journey Time (SJT)
Not realistic
Route In-vehicle Waiting Walking Total
[min] [min] [min]
Direct 20.5 5.60 3.60 29.70
Indirect 20.5 2.94 3.69 27.12
Nadudvari, Liu, HessITS Uni of Leeds
Bayesian Modelling Framework (BMF)
UTSG 2015, City University London
06 January 2015
ZJT frequencies
ZJT [min]
Oys
ter
tran
sact
ion
s [#
]
Supposing one route
Setting a Gaussian distribution
Calculating parameters
Comparing with Scheduled Journey Time (SJT)
Mean = 27.84 minSD = 4.03 min
Route In-vehicle Waiting Walking Total
[min] [min] [min]
Direct 20.5 5.60 3.60 29.70
Indirect 20.5 2.94 3.69 27.12
Nadudvari, Liu, HessITS Uni of Leeds
Random Utility Maximisation (RUM)
UTSG 2015, City University London
06 January 2015
RAVEAU, S., GUO, Z., MUÑOZ, J. C. & WILSON, N. H. M. 2014. A behavioural comparison of route choice on metro networks: Time, transfers, crowding, topology and socio-demographics. Transportation Research Part A: Policy and Practice, 66, 185-195.
Parameters already calibrated for the London Underground.
Time Transfer Crowding Topology Socio-demographics
In vehicle time is identical for two routes as it is a common line problem
In vehicle Wait Walk
Time
In vehicle
Nadudvari, Liu, HessITS Uni of Leeds
Calibrated parameters in Raveau 2014
RUM
Random Utility Maximisation (RUM)
UTSG 2015, City University London
06 January 2015
Time Transfer Crowding Topology Socio-demographics
In vehicle Wait Walk
Further research:• “A”: Headway: Two service, same platform?• “B”: Wait time for infrequent service? • “C”: Arrival from “A”. Services on time?
Now we consider: wait time = headway/2• “A”: 2.88/2=1.44 min• “B”: 11.20/2=5.60 min →• “C”: 3.00/2=1.50 min
Parameter: θwait=-0.269-0.208=-0.477
Morden
South Wimbledon
Colliers Wood
Tooting Broadway
Tooting Bec
Clapham Common
Balham
Clapham South
Clapham North
Stockwell
Oval
Waterloo
Tottenham Court Road
Embankment
Charing Cross
Leicester Square
Goodge Street
B
B
B
A
C
C
Route 1: 5.60 min
Route 2: 2.98 min
Defaultvalue
Adjustment for AM peak
Applied value
Nadudvari, Liu, HessITS Uni of Leeds
Random Utility Maximisation (RUM)
UTSG 2015, City University London
06 January 2015
Time Transfer Crowding Topology Socio-demographics
In vehicle Wait Walk
Departure/ Arrival same platform → Access and Egress times identicalInterchange: adjacent platforms → Short interchange time: 0.09 min.
Time Transfer Crowding Topology Socio-demographics
Crowding not known → Depends on the RC of other OD pairs → RCs are not independent of each other →Not only single RC problems for OD pairs →Transit Assignment model for a network.
Common line → Identical topological perceptions
Crowding Topology
Nadudvari, Liu, HessITS Uni of Leeds
Random Utility Maximisation (RUM)
UTSG 2015, City University London
06 January 2015
Utility
𝑃𝑖 =𝑒𝑈1
𝑒𝑈1+𝑒𝑈2=
𝑒−2.671
𝑒−2.671+𝑒−2.138=37%
𝑈1 = 𝑇𝑤𝑎𝑖𝑡,1 ∙ 𝜃𝑤𝑎𝑖𝑡 + 𝑇𝑤𝑎𝑙𝑘,1 ∙ 𝜃𝑤𝑎𝑙𝑘 +∙ 𝜃𝑇𝑅,1 =
5.60 ∙ −0.477 + 0 + 0 = − 2.671
𝑈2 = 𝑇𝑤𝑎𝑖𝑡,2 ∙ 𝜃𝑤𝑎𝑖𝑡 + 𝑇𝑤𝑎𝑙𝑘,2 ∙ 𝜃𝑤𝑎𝑙𝑘 +∙ 𝜃𝑇𝑅,2 =
2.98∙ −0.477 +0.09∙ −0.323 + (−0.708) = − 2.138
Route Choice Probability
Direct route
Indirect route
Direct route
31 % from RODS
Nadudvari, Liu, HessITS Uni of Leeds
UTSG 2015, City University London
06 January 2015
Random Utility Maximisation (RUM)
Morden
South Wimbledon
Colliers Wood
Tooting Broadway
Tooting Bec
Clapham Common
Balham
Clapham South
Clapham North
Stockwell
Oval
Waterloo
Tottenham Court Road
Embankment
Charing Cross
Leicester Square
Goodge Street
Indirect route: Save 2.6 min
𝑃𝑖 =𝑒𝑈1
𝑒𝑈1 + 𝑒𝑈2=
1
1 + 𝑒𝑈2−𝑈1
Morden – Goodge Street: 41,4 minOval – Waterloo: 14,9 minPerceived same to save 2.6 min for 2 cases?
Cost dampingDALY, A. 2010. Cost Damping in Travel Demand Models - Report of a study for the Department for Transport. RAND Europe.
Probability from utility difference
𝑇𝑤𝑎𝑖𝑡,1 − (𝑇𝑤𝑎𝑖𝑡,2−𝑇𝑤𝑎𝑙𝑘,2) =
5.60-(2.98-0.09)=2.6 min
Nadudvari, Liu, HessITS Uni of Leeds
Conclusions and further research
UTSG 2015, City University London
06 January 2015
• Zone to zone OD pairs → Larger dataset, better for analysis
• Bayesian Modelling Framework (BMF)• Observed data to infer route choice
• If 2 routes similar OJT, mixture of 2 comp. not fit well, 1 fits better
• Random Utility Maximisation (RUM)• Scheduled data to estimate route choice
• Interdependence of crowding → Not only RC model for OD pairs, but TAM for network
• Considers only the utility difference → Cost damping
Nadudvari, Liu, HessITS Uni of Leeds
Conclusions and further research
UTSG 2015, City University London
06 January 2015
• Combination of BMF and RUM• Observed AND scheduled data to have a better picture of route choice
• Infer service taken from entry/exit time and departure/arrival time
• Passenger arrival and preference behaviour• Arrive randomly or before the departure of service?
• Wait for preselected service or board first arriving service?
Nadudvari, Liu, HessITS Uni of Leeds
References
• CHAN, J. 2007. Rail Transit OD Matrix Estimation and Journey Time Reliability Metrics Using Automated Fare Data. Master of Science in Transportation, Massachusetts Institute of Technology.
• FU, Q., LIU, R. & HESS, S. 2014. A Bayesian Modelling Framework for Individual Passenger’s Probabilistic Route Choices: a Case Study on the London Underground. Transportation Research Board (TRB) Annual Meeting.
• DALY, A. 2010. Cost Damping in Travel Demand Models - Report of a study for the Department for Transport. RAND Europe.
• RAVEAU, S., GUO, Z., MUÑOZ, J. C. & WILSON, N. H. M. 2014. A behavioural comparison of route choice on metro networks: Time, transfers, crowding, topology and socio-demographics. Transportation Research Part A: Policy and Practice, 66, 185-195.
• SCHMÖCKER, J.-D., FONZONE, A., SHIMAMOTO, H., KURAUCHI, F. & BELL, M. G. H. 2011. Frequency-based transit assignment considering seat capacities. Transportation Research Part B: Methodological, 45, 392-408.
• SUN, L. 2014. Characterizing Travel Time Reliability and Passenger Path Choice in a Metro Network. Paper presented at the hEART (European Association for Research in Transportation ) Conference, 10-12 September 2014,. Leeds.