public transport in the era of maas - delft university of...

Public Transport in the era ofMaaSdr. Konstantinos GkiotsalitisCenter for Transport Studies

University of Twente — Quick Facts

I Is highly specialized in engineering.

I >10,000 students & 1,000 successful ventures to date.I Collaborates with Delft University of Technology,

Eindhoven University of Technology and theWageningen University and Research Centre under theumbrella of 4TU.

Public Transport in the era of MaaS October 9, 2018 2 / 30


I Is highly specialized in engineering.I >10,000 students & 1,000 successful ventures to date.

I Collaborates with Delft University of Technology,Eindhoven University of Technology and theWageningen University and Research Centre under theumbrella of 4TU.



I Is highly specialized in engineering.I >10,000 students & 1,000 successful ventures to date.I Collaborates with Delft University of Technology,

Eindhoven University of Technology and theWageningen University and Research Centre under theumbrella of 4TU.


Location


UT Center for Transport Studies

StructureI Faculty of Engineering Technology

I Department of Civil EngineeringI Center for Transport Studies


My role at UT

EducationI Public Transport Modelling (course leader)I Traffic Operations (course leader)I Network Modelling (lecturer)I Mathematical Optimization (lecturer)

ResearchI Public Transport Planning (tactical and operational level)I Demand Prediction with new data forms (social media,

cellular data)I Traffic Operations (esp. connected vehicles)


Before that...

Research Associate - Research ScientistPT Customers: Singapore, Tokyo, Hong Kong, India, Santiago

4 University of Twente

Quick Facts

5-year diploma

MSc

PhD

8 Filed Patents

Laboratories EuropeIntelligent Transport Systems

Data Science Applications2012-2018 3 Technology

Transfers

11 FP7/H2020 Proposals


Differences in New Life...

Business Forces

Area Business Model Impact

Singapore,London, HongKong

Almost every yearanother set of buslines is open forbidding

Strong competition onimproving operationalKPIs and upgrading ICT

Twente (mostly inthe Netherlands)

winner-takes-it-allcontract for ∼10yrs

1. PTOs want to identify& terminatenon-profitable lines2. Market share ischallenged from mobilitystart-ups


Biggest PTO problems

Make a guess...

I Bus drivers call in sick (rate way above the nationalaverage)

I Demand changes over the years and some lines are notprofitable

I Passengers tap off some stops before arriving at theirdestination. Why?

I Fear or losing market share from mobility start-upsI These are real problems. Well, if we ask UITP we will see

adopting AI and other trends as No1 issue



Make a guess...I Bus drivers call in sick (rate way above the national

average)

I Demand changes over the years and some lines are notprofitable







average)I Demand changes over the years and some lines are not

profitable








profitableI Passengers tap off some stops before arriving at their

destination. Why?








destination. Why?I Fear or losing market share from mobility start-ups

I These are real problems. Well, if we ask UITP we will seeadopting AI and other trends as No1 issue






destination. Why?I Fear or losing market share from mobility start-upsI These are real problems. Well, if we ask UITP we will see



Mid-term Research Area

Top-down Planning of MaaS / Seamless integration with PTNowI PTO terminates bus line services with low demand (mostly

periurban) areasI ”Old” buses are provided to volunteers who can facilitate

those lines at will

FutureI PTO uses more flexible schemes (interlining,

short-turning) to reduce the number of terminated servicesI unprofitable services are substituted from PTO-driven

MaaS with a guaranteed regularity and fare price (i.e.,KEO-bike)


Mid-term Research Area

Top-down Planning of MaaS / Seamless integration with PTNowI PTO terminates bus line services with low demand (mostly

periurban) areasI ”Old” buses are provided to volunteers who can facilitate

those lines at willFutureI PTO uses more flexible schemes (interlining,

short-turning) to reduce the number of terminated servicesI unprofitable services are substituted from PTO-driven

MaaS with a guaranteed regularity and fare price (i.e.,KEO-bike)


Research Challenges

I Flexible Bus PlanningI Determination of (persistently) unprofitable linesI Top-down MaaS modelling and service take-up

(gamification, incentives, accessibility studies [H2020])

Lines of ResearchI Large-scale Optimization, Operational ResearchI Demand PredictionI Accessibility and Behavioral Analysis


H2020 Calls

I H2020 MG-4-5-2019: An inclusive digitally interconnectedtransport system meeting citizens’ needs (1 stage, April25th 2019)

I LC-MG-1-2-2018: Spatial planning implications ofRegional multimodal Mobility Innovations in Europe

I Netherlands Organisation for Scientific Research (NWO)?


Selected Recent Research

Public Transport (2018)I K.Gkiotsalitis, N.Maslekar (2018). Towards transfer synchronization of regularity-based bus operations with

sequential hill-climbing, Public Transport 10(2), 335-361.I K.Gkiotsalitis, O.Cats (2018). Reliable frequency determination: Incorporating information on service

uncertainty when setting dispatching headways, TR Part C: Emerging Technologies, 88, 187-207.I K.Gkiotsalitis, O.Cats (2018). A cost minimization model for bus fleet allocation incorporating the

generation of short-turning and interlining options, TR Part C: Emerging Technologies, 2nd stage review.I K.Gkiotsalitis, N.Maslekar (2018). Multiconstrained timetable optimization and performance evaluation in

the presence of travel time noise, ASCE Journal of Transp. Engineering Part A Systems, 144(9),04018058.

Demand PredictionI K.Gkiotsalitis, A.Stathopoulos (2016). Joint leisure travel optimization with user-generated data via

perceived utility maximization, TR Part C: Emerging Technologies, 68, 532-548.I K.Gkiotsalitis, A.Stathopoulos (2015). A utility-maximization model for retrieving users’ willingness to travel

for participating in activities from big-data, TR Part C: Emerging Technologies, 58, 265-277.I K.Gkiotsalitis, A.Stathopoulos (2016). Demand-responsive public transportation rescheduling for adjusting

to the joint leisure activity demand, International Journal of Transportation Science and Technology,5(2), 68-82.

Traffic TheoryI AHF Chow, Y. Li, K.Gkiotsalitis (2015). Specifications of Fundamental Diagrams for dynamic traffic

modeling, ASCE Journal of Transportation Engineering, 141(9), 04015015.


Ongoing Research

TECHNICAL TOPIC: Robust Network-wide synchronizedscheduling of public transport services


Objective of this work

Stops of line 1

Stops of line 4

Bus Lines

Transfer stops

Essingetorget

Frihamnen

Gullmarsplan

Radiohuset

Common bus stops

Bus lines 1 and 4 in Stockholmwith 5 transfer stops

Solving the multi-line synchronization problem of busservices at the tactical planning stage considering:

I the variability of the travel and dwell times of bus trips;I the regularity of individual bus lines;I and the operational regulatory constraints related to the

schedule sliding prevention; and the layover time limits.



Stops of line 1

Stops of line 4

Bus Lines

Transfer stops

Essingetorget

Frihamnen

Gullmarsplan

Radiohuset

Common bus stops


Solving the multi-line synchronization problem of busservices at the tactical planning stage considering:I the variability of the travel and dwell times of bus trips;

I the regularity of individual bus lines;I and the operational regulatory constraints related to the




Stops of line 1

Stops of line 4

Bus Lines

Transfer stops

Essingetorget

Frihamnen

Gullmarsplan

Radiohuset

Common bus stops


Solving the multi-line synchronization problem of busservices at the tactical planning stage considering:I the variability of the travel and dwell times of bus trips;I the regularity of individual bus lines;

I and the operational regulatory constraints related to theschedule sliding prevention; and the layover time limits.



Stops of line 1

Stops of line 4

Bus Lines

Transfer stops

Essingetorget

Frihamnen

Gullmarsplan

Radiohuset

Common bus stops


Solving the multi-line synchronization problem of busservices at the tactical planning stage considering:I the variability of the travel and dwell times of bus trips;I the regularity of individual bus lines;I and the operational regulatory constraints related to the



Bus Movement in the presence of uncertainty

s=3

bus stop s=1 s=2

s=4

n-th bus trip of the day

Bus line l

𝑡𝑙,𝑛,𝑠=2 𝑘𝑙,𝑛,𝑠=2

Which is the arrival time of bus trip n at stop s if it is dispatched attime xl,n?

It is given by: al,n,s = xl,n+s∑

z=2

(tl,n,z + ξl,n,z)+s−1∑z=1

(kl,n,z + ζl,n,z) (1)


Mathematical Program of the Network-wideSynchronization

Objective: Improve Regularity of individual services (solutionthat performs best at its worst-case noise scenario)

Constraints: Transfer Synchronization; Avoid Schedule Sliding

Decision Variables: Dispatching times of all bus trips

(Q) : minx

maxξ,ζ

f (x , ξ, ζ) (2)

s.t.: x ∈ F(ξ, ζ) ={

x x satisfies the constraints}

(3)

xl,1 = δminl , ∀l ∈ L (4)

ξminl,s ≤ ξl,n,s ≤ ξmax

l,s , ∀l ∈ L, ∀n ∈ N(l), ∀s ∈ S(l) \ {1}(5)

ζminl,s ≤ ζl,n,s ≤ ζmax

l,s , ∀l ∈ L, ∀n ∈ N(l), ∀s ∈ S(l) (6)


Infeasibility problem

If the travel and dwell times can take any value resulting toextreme-case scenarios there might be no dispatching timesolution that:I satisfies the schedule sliding constraints;I ensures that all trips at transfer stops are synchronized.

We thus introduce a penalty function f̃ (x , ξ, ζ) that penalizesthe objective function f (x , ξ, ζ) if some constraints are notsatisfied.


Case Study

Stops of line 1

Stops of line 4

Bus Lines

Transfer stops

Essingetorget

Frihamnen

Gullmarsplan

Radiohuset

Common bus stops


Indication of Search for a Robust Solution

1 5 10 15 20 25 30Iterations

0.6

0.8

1.0

1.2

1.4

1.6

Valu

e of

the

pena

lized

obj

ectiv

e fu

nctio

n (s

ec2 )

1e10

Bounded worst-case performance of most robust solutionSolution performance on worst-case noisePerformance of new solution subject to the previous worst-case noise

Figure: Convergence of the alternating optimization. The robust solutionreduces the worst-case penalized objective function value in worst-casescenarios by ∼60% from 1.727E+10 to 0.701E+10


Main Dilemma

Case 1: Robust synchronization to common-case scenarios(travel time and dwell time deviations of up to 10%) might notbe robust enough to extreme cases

Case 2: Robust synchronization to extreme scenarios (i.e.,travel and dwell time deviations of up to 30% from theirexpected values) might perform badly during normal days

⇒ find the sweet spot of each case study


Observations using 30 days of data

Average Excessive Waiting median max.Time per passenger (min) min. Q1 median Q3 max. impr. impr.

Originally Planned Schedule 1.443 1.583 1.626 1.673 1.810 N/A N/A

Robust to 10% deviations 1.372 1.495 1.542 1.587 1.710 5.17% 5.52%

Robust to 30% deviations 1.442 1.622 1.654 1.661 1.6811 -1.72% 7.13%

Average Waiting Time for median max.Transferring Passengers (min) min. Q1 median Q3 max. impr. impr.

Originally Planned Schedule 1.888 2.229 2.381 2.579 2.910 N/A N/A

Robust to 10% deviations 1.511 1.692 1.710 1.984 2.811 28.18%2 3.40%

Robust to 30% deviations 1.819 2.231 2.403 2.468 2.581 -0.92% 11.31%3

1service regularity will remain very close to the median value in extreme cases when using an ”overprotective”

timetable2

extremely strong improvement of the median of the waiting times of transferring passengers when planningfor 10% deviations

3waiting times of transferring passengers will remain very close to the median value in extreme cases when

using an ”overprotective” timetable (the same improvement was only 3.40% when planning for 10% deviations)


Ongoing Research

TECHNICAL TOPIC: Bus fleet allocation incorporating thegeneration of short-turning and interlining options


Low Bus Occupancy problem

1 5 10 15 20 25 30 35Bus stops

0

20

40

60

80

100

Ave

rage

load

(pas

seng

ers/

h)

72% drop

Bus line 3

Inbound directionOutbound direction

1 5 10 15 20 25 30 35Bus stops

0

25

50

75

100

125

150

Ave

rage

load

(pas

seng

ers/

h)

27% drop

Bus line 5

Inbound directionOutbound direction

Figure: Average bus-load per bus stop for bus line 3 and bus line 5 inThe Hague from 4 pm to 5 pm


Rule-based split of original lines

Line 1

Line 2

Switch

Stops

Bus Stops

Short-turn line

Interlining

line

Figure: Potential generation of short-turning lines (blue dashed) orinterlining lines (red) at specific switch stops (orange)


Rule-based split of original lines

Line 1

Line 2

Line 3

Line 5

Line 6

Line 8

Bus stop 1 Bus stop 6

Bus stop 12

Figure: Example of rule-based generation of all possible switch stopsfor bus line 3 in Den Haag


Example of Implementation in Den Haag

Line 1

Buses: 21

Line 2

Buses: 19

Line 3

Buses: 14

Line 4

Buses: 22

Line 5

Buses: 22

Line 6

Buses: 21

Line 7

Buses: 38

Line 8

Buses: 6

Short-turn Line 1; Buses: 5

Short-turn Line 3

Buses: 4

Interlining 1<->5

Buses: 8

Short-turn Line 2;

Buses: 6

Interlining 2<->4

Buses: 4

Interlining 7<->6

Buses: 10

Figure: Original bus network in Den Haag and allocation toshort-turning and interlining lines


Theoretical Gain

0

10000

20000

30000

40000Co

sts (

euro

s)43436

41445

Waiting time costs

21616

18621

Vehicle running costs

3980 4000Depreciation costs

Originally planned linesAll lines

60000

62000

64000

66000

68000

70000

72000

Tota

l Cos

ts (e

uros

)

69032

64066

Originally planned linesAll lines

Figure: Costs when using (a) originally planned lines only and (b)originally planned lines along with interlining and short-turning lines


But what will happen if we need more practicalschedules?

60%70%

80%90%

100%

z10%

20%30%

40%50%

Pena

lty fu

nctio

n (e

uros

)

64000

65000

66000

67000

68000

69000

Figure: Total Cost changes: (i) ψ controls the minimum percentage ofbuses that should be allocated to originally planned lines; (ii) zcontrols the minimum ridership change from stop to stop that justifiesthe generation of a switch stop


New Research Items

I Correlation of Driver Behavior with Traffic Density (PhDCandidate Emiliano Heyns)

I Real-time Bus Holding with Reinforcement Learning (DrFrancesco Alesiani)

I Forecasting spatio-temporal variations in ODs of bikesharing systems (MSc Jan M. Engels - Yellowbike)

I Dynamic optimization of train allocation to a trainyard(MSc Bram Schasfoort - ARCADIS)

I Scheduling of Electric Buses (Dr Guelcin Emris)I Increase MaaS inclusion by studying user-groups with the

use of smartcard, cellular, smartphone App data andsurveys (Prof Karst Geurs)


Thank-you

Thank-you for your time!

E: [email protected]: +31534891870W: https://people.utwente.nl/k.gkiotsalitis?tab=about-me


public transport in the era of maas - delft university of...

Documents