Avoiding bus bunching: From theory to practice

AvoidingBusBunching:FromTheorytoPractice

[email protected]

Department ofTransport EngineeringandLogisticsPontificiaUniversidadCatólicadeChile

Jointworkwith:JuanCarlosMuñozandFelipeDelgado

Outline

• Thebusbunchingproblem

• Proposedcontrolstrategy

• Simulationresults

• PilottestresultsandImplementationChallenges

• Conclusions

Motivation:Efficiencyintheuseofroadspace

www.BRT.cl

Whatcanwesayabouttheuser?

• Perceiveswaitingtimeandwalkingtimetwiceasimportantastraveltimeinsidethevehicle.

• Avoidstransferring,speciallyiftheyareuncomfortable

• Needsareliableexperience

• Requestsaminimumcomfortexperience

• Requestsinformation

• Needstofeelsafeandsecure

Whatcanwesayaboutbusservice?

Busiscriticaltoprovideagooddoor-to-doortransitalternativeformanyjourneys:

• Muchhighernetworkdensityandcoveragethanrail

• Greaterflexibilityinnetworkstructure

• Lowmarginalcostforserviceexpansion

BUT as traditionally operated, it also has serious limitations:

• Low-speed• Subject to traffic congestion• Unreliable• Harder to convey network to the public• Negative public image

MetroAttributes

Actions

IncreaseSpeed

Regular Headways

Maindrivers

IncreaseCapacity

IncreaseFrequency

•Segregated ways/lanes•Reducedwell times

•Farepaymentoff-bus•Buseswithmultipledoors

•Increase distance between stations•Expressservices•Trafficsignalpriorityandpriorityatintersectons•Improvedheadwayregularity

Fast ComfortLow waits Reliable

RicardoGiesen©

Bus

Bus

Stop Stop

Waiting Passengers

WaitingPassengers

Bus Operations without Control

RicardoGiesen©

BusBusStop Stop

a small perturbation…

Waiting Passengers

WaitingPassengers


RicardoGiesen©

Bus

Bus

Stop Stop

While one bus is still loading passengers the other bus already left its last stop


RicardoGiesen©

Bus

BusStop Stop


RicardoGiesen©

BusBus

Stop Stop

Without bus control, bus bunching occurs!!!


Transantiago,Santiago,Chile

Cambridge,MA

Time-spacetrajectoriesLine210

Busbunchingisasevereproblem

§ Most passengers wait longer than they should for crowded buses

§ Put pressure in the authority for more buses

§ Reduces reliability affecting passengers and operators

Idea:ControlStrategiestoAvoidBusBunching!

SolvingBusBunching:ScheduleControl

• Buseshavetoadheretostaticschedulesinstopsorcontrolpoints

• Usefulforlowfrequency(<5buses/hour)lineswithoutcongestion:nottherealityofmajorurbancities.

• KPI:• Punctualityindicatorthatmeasurestheadherencetoaschedule.

• Pros:• Easytoimplement&tounderstandbydriver.

• Transparentforusers.

• Cons:• Notusefulforhighfrequencyorcongestedlines

• Excessiveslackinschedules

• Dynamicsuggestionstoincrease/decreasespeed&holdingatstopsandterminals

• Severalcontrolstrategiesintheliterature,latelysomedeveloped&implemented:

• RecentadvancementsonICTs

• Lowerhardware&communicationcosts

• Challenges:• Makingasmartsolutionrobustenoughtoscale

• Drivercompliance

• KPI:• Headwaydeviationincontrolpointsalongtheroute

• Waitingtimes,Busloads

• Santiago:ICR&regularityfines

SolvingBusBunching:HeadwayControl

PreviousWorkonHoldingStrategies

Reference PH PD and RT Overtaking OF Veh. Cap. Control Points Buses Sol. method

Ding y Chien (2001) Multiple Deterministic Prohibited Vh ignored CMS One OPT

Eberlein et al (2001) Multiple Deterministic Prohibited Wfirst ignored PSS Multiple Heuristic

Hickman (2001) One Stochastic Allowed Wfirst+Win-veh ignored PSS One OPT

Zhao et al (2003) One Stochastic Prohibited Wfirst+Win-veh ignored CMS One Heuristic

Sun y Hickman (2004) Multiple Deterministic Prohibited Wfirst+Win-veh ignored PMS Multiple Heuristic

Zolfaghari et al (2004) Multiple Deterministic Prohibited Wfirst+Wextra considered CSS Multiple Meta-heuristic

Puong y Wilson (2004) Multiple Deterministic Prohibited Wfirst+Win-veh+Wextra considered CMS Multiple OPT

NewIdea:BoardingLimits

Turnquist and Blume (1980) pioneer work on threshold policy for holding buses

Innovations

1.- Decisionvariables:HoldingBoardingLimits

Passengers allowed to board

Passengerspreventedfrom

boarding

This can be used even when at less thanphysical capacity in order to increase operatingspeed.

2.- Buscapacityincorporatedwithoutresortingtobinaryvariables

SystemCharacteristics

Thesystemiscomposedby:

One-wayloopTransitcorridor.

Operatedbyasingleservice.

N stops.

K homogeneousbuses.

StateVariables

Real Time information about:

Bus position.

Bus loads.

# of Passengers waiting at each stop.

However,wecouldworkwithestimations…

andinpracticethereisnoalternative

Model:Assumptions

Someinformationabouttripdestinations.

Dwelltime:dominatedbyboardings.

Busesserveallstopsandovertakingisnotallowed.

Every time a bus reaches a stop:

How much to hold it?

Should we prevent some passengers from boarding?

Solve a rolling horizon optimization problem to take those single decisions

1

2

N/2

N-1NCurrentSituation

Busarrives atastop

Model:Problemdefinition

1

2

N/2

N-1NCurrentSituationOptimizePredicting






1

2

N/2

N-1NCurrentSituation

Apply decision






3. Model: Objective Function

1 2 3 4,kn knfirst in veh extrah w

Min W W W PEq q q q-× + × + × + ×

Experiment:SimulationScenarios

Highfrequencytransitsystemscenario(headway~2min.)

Highpassengerdemand(buscapacitiescanbereached)

One-wayloopTransitcorridorwith30Stopsand29TrafficLights(inbothways)

Competingflow700veh/h

Mainflow900veh/h

Saturationflow1800veh/h

Parameter ValueCycletime 72 sec

Greentimeforbusapproach 40 sec

Offsetsbetweentrafficlights 23 secMaximumextension 4 sec

Nocontrol

Spontaneousevolutionofthesystem.

Busesaredispatchedfromtheterminalassoonastheyarriveoruntiltheyreachthedesignedheadway.

Noothercontrolactionsaretakenalongtheroute.

ProposedModel

Solvetherollinghorizonoptimizationmodelincludingholding,boardinglimitsandgreenextension

Experiment:SimulatedStrategies

5. Simulation Results

30 runs for every combination of strategies and scenarios

Each run represents 2 hours of bus operation.

15 minutes “warm-up” period.

Variability is introduced in the simulation experiment.

Results:SimulationFramework

Simulator AMPL Solver

State variables

ControlAction

NoControl ProposedWfirst 7636.32 1438.62Std.Dev. 649.36 146.56%reduction -81.16Wextra 6218.71 1010.52Std.Dev. 2265.24 82.04%reduction -83.75Win-veh 175.32 1561.34Std.Dev. 31.69 77.3%reduction 790.55Wt_light 4052.81 2965.1Std.Dev. 88.27 110.27%reduction -26.84Total 18083.16 6975.58Std.Dev. 2600.63 275.12%reduction -61.42

SimulationResults:TransitUsers

SimulationResults:BusLoads

72

SimulationResults:CycleTimes

Results:Vehiclecap.constraints&mediumfrequency

0-5min. 5-10min. morethan10min.Nocontrol 78.90 17.52 3.58Tresholdcontrol 89.26 9.80 0.95HRT 92.46 7.50 0.04HBLRT 93.74 6.19 0.07

%ofpassengersthathavetowaitbetween:

ResearchConclusions

– We have a tool for effectively reducing bunching of buses in a BRT.

– The tool is fast enough for real-time applications.

– The proposed control strategy outperform simple control ruleswith saving up to 61.4% for transit users.

– Boarding Limits are only attractive in high demand and highfrequency scenarios.

– Severely improve comfort and reliability for transit users.

– Reduction on vehicle cycle times allow for reductions on fleet sizeor improvements on level of service.

Weshoulddoit!

PilotProjectContext

• In2012afineschemewasimplementedintheSantiagotransitsystem(Transantiago):üGuaranteefrequencyandregularitycompliance

üRegularityfinesifbusheadwayexceedsthreshold

üAveragemonthlysystemfines(2012):$USD2.3million

• Thefineschemeevidencedthelackofavailabletoolsinthemarkettoregularizeheadwaysonhighfrequencyservices

Howdoesthemodeloperate?

Busservice

t

Servicestartsoperating

t=0

t=to

Optimizationmodel

Datacollection Instructionstobusoperator

Holdingtime/Departure

time

Current serviceconditions (buspositions,etc)

t=2to

Timeintervalbetweenoptimizations

Optimizationmodel

Holdingtime/Departure

time

Current serviceconditions (buspositions,etc)

Pilotstudies

• Duetothesuccessofthemodelinsimulationenvironment,aSantiagobusoperatorcompanywaswillingtotry.

• Pilotstudieswithservice210(SubusChile):

üNovember2012(Pilot1)andApril2013(Pilot2)

OptimizationModel

Monitorsat24busstop

ControlInstruction:Busdeparturetime

TextmessagingBusdriver

Busdeparturetime

Directcommunication

210PilotPlan

• Exampleofbusbunchingin210service:

Service210Characteristics

• Highdemand(48,000passengers/day)

• Highfrequencyservice(abusevery3-4minutes)

• Completeroute(inbound+outbound):135stopsand56kmlong

• Inmorningpeak:upto60busesoperatinginservice

• Oneoftheworstserviceintermsofuser evaluationofTransantiago

• Servicewithhighfareevasion

Pilotstudies

Pilotstudies:results• Morningpeakperiod(6:30-9:30)fines210-outboundservice($CLP):

$10,000

$30,000

$50,000

$70,000

$90,000

$110,000

USD1=CLP670

Pilotstudies:results

Bustrajectories (distance vstime)day without control

Pilotstudies:results

Pilotstudies:resultsLine210

• Surprisingly,uservalidations(demand)increasedby20%duringthepilotplans

– Reductionofuserevasion:“passiveevasion”

• Nosignificanteffectsonbusfrequencyandcycletimes

INNOVATIVE TECHNOLOGICAL SOLUTIONS FOR TRANSIT SYSTEMS

WWW.TRANSITUC.COM

FelipeDelgado,RicardoGiesen,JuanCarlosMuñoz

PedroLizana

Headwaycontrolsoftware

GPSdata:buslocation

BuslocationWebservice

BusOperator/Agency

Controlinstruction

Webservice

GPSdevice

Controlinstruction

User smartphone AppOnlinevisualization tool

APCdata:demanddata

Real-timebustraveltimepredictor

Datacollectionsmartphone App

Demand data

Transit UC

TechnologicalPilotPlans

• Objective:Sendcontrolinstructionsdirectlytobusdriverusingabusconsole-tablet

• Subus-Chile(210service)

üControlinstructions:holdingbusesatstopsandskipstops

üTheyalreadyhaveindustrialconsolesinstalledintheirbuses:usedwithaitinerary-basedcontrolsystemthathasbeenineffectiveinpeakperiods

Softwareinputinformation• Statictransitsystemdata:

– Busservices,operatingprograms,busstoplocations,etc.:datainformatGeneralTransitFeedSpecification(GTFS)

• Real-timebuspositions:– GPSalreadyinstalledinbuses– TabletAppGPS

• Demanddata:– Passivesmartcardinformation:ODmatricesandbusstoparrivalrates

• Segmentspeeds:– Combinationofreal-timespeeds(fromGPSdata)withhistorical

speeds:researchonArtificialNeuralNetwork(ANN)andBayesianNetworksalgorithms

210PilotPlan– Busconsole


• Screenshotofholdinginstructionatbusstop:

Holdingtimecountdown

Currentstop

Copyrights:Mobius

• Screenshotofholdinginstructionnotcarriedoutbybusdriver:


Helpamoreregularoperation:departstopatindicatedtime

Copyrights:Mobius

210PilotPlan:TechnicalChallenges

• SomebuseshaveGPSdatadelayedinupto3minutes(withaGPSpulseevery30seconds):Hardtopredictwherethebusiscurrentlylocated.

• VeryrigidconsoleApp(developedbyothercompany):insomecasestheconsoledoesnotconnectandnosugestionsaredisplayedtodrivers.

210PilotPlan- HumanChallenges

• Somedriversliketobunchupwithpreviousbusessofewpeopleloadtheirbusesandworkshortershifts

• Somedriversbandalizedtheirconsoles:companyhadtoputaplasticlayerontopofthescreen(notanymoretouchscreen)

• Usersfrustatedbythecurrentlevel-of-servicethatdonotliketobeheldatbusstops(driversthreatenedtobebeatenbyusers;userskickingbusandwindows;etc)

210PilotPlan:CompanyChallenges

• Thecompanyhasanexistingschedule-basedcontrolsystem(withverylowcredibilitybetweendrivers)thatinteractswithourheadway-basedsystem.

– Possibleconfusioninformationtodrivers

• Companydidnotprovideanyincentivesfordrivertoimproveregularity.

PedroLizana,CEOTransitUC

RicardoGiesen

RESULTSPILOTTESTTRANSMILENIO:DUAL84

Service84,Transmilenio

Howtomeasureregularity?

• Waitingtimeatstopscanbeexpressedas:

• Thefirsttermdependsonthefrequencyofbuses(1/H),whichisafunctionofthenumberofbuses(n)andthecycletime(Tc).

H=Tc/n

• Thesecondtermdependsonthevariabilityoftheheadwaysbetweenbuses.ThiscanbeimprovedusingBuzzAssist.

E(W ) = E(H2 )

2E(H )=E(H )2

+Var(H )2E(H )

Transmilenio PrePilot Results

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0

3% 3% 13% 13% 23% 23% 33% 33% 43% 43% 53% 53% 63% 63% 73% 73% 83% 83% 93% 93%

Pre-Pilot1:3/16withControlvs.3/17withoutcontrol

Var.sincontrol

Fijosincontrol

Var.concontrol

Fijoconcontrol

ControlPoint

Waitin

gtim

e[m

in]

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0

3% 3% 13% 13% 23% 23% 33% 33% 43% 43% 53% 53% 63% 63% 73% 73% 83% 83% 93% 93%

PrePilot2:4/14with Controlvs.4/21withoutcontrol

Var.sincontrol

Fijosincontrol

Var.concontrol

Fijoconcontrol

ControlPoint

Waitin

gTime[m

in]

Transmilenio PrePilot Results

140

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

3% 3% 13% 13% 23% 23% 33% 33% 43% 43% 53% 53% 63% 63% 73% 73% 83% 83% 93% 93%

Pilot:5/26withControlvs.5/25withoutcontrol

Var.sincontrol

Fijosincontrol

Var.concontrol

Fijoconcontrol

ControlPoint

Waitin

gTime[m

in]

Transmilenio PilotResults

Results:CycleTime(onedirection)

141

0

10

20

30

40

50

60

70

80

90

100

Cycle

Tim

e(m

in)

TpoCicloconControl(14Abril+26Mayo)

TpoCiclosinControl(21Abril+25Mayo)

With control: Average Cycle Time=63.4minandSt.Dev.=5.6minWithoutcontrol: Average Cycle Time=69.5minandSt.Dev.=11.8min

Promisingresults

• Moreregularity

• Lessfines

• Moredemand

• However…driversdonotfollowinstructions,implementationchallenges…

AVOIDING BUS BUNCHING

DevelopedAppforIndustrialConsoletoProvideDriverAssistance

ImplementationinRedbus,Transdev

• 61linesand600+buses

• Approx.300,000dailypaidtrips(bip!)with~ 25%fareevasion

0

1

2

3

4

5

6

7

0 5 10 15 20 25

Fine

s

Distance from origin [km]

Regularityfinesforeachbusoperatorcompany(DICTUC,2016)

U1

U2

U3

U4

U5

U6

U7

Dispatchingatterminalsisnotregular

SynopticFleetcontrol&monitoring

CommercialDispatcherComplywithfrequency,regularityandindicatorsSmartdispatcher

TerminalDispatcherOpening&closingofdrivershiftsNoncommercialmovements

ConsoleDriverassistanceOn-routeregularity

TrainingDispatchers & drivers

DemandPerstop&lineLoadprofileFareevasionsupervision

UsersReal-timeupdatestousers

SurveyingPassengercountingApp

Regularity

Regularity

Evasion

• Controlcenteroperatorshaveaccesstoalivesynoptictovisualizebuses,theirbusbunchingavoidanceinstructions,changetheirstatusandsendmessagestodriveranddispatchers.

CAD/AVL

• DispatchershaveaccesstoawebandAndroidmobileappDispatchApp,thatinformsthemofreal-timeregularityindicatorsandgivesthemsmartdispatchtimesuggestionsinordertoincreasetheregularityofeachline.

Smartdispatcher

• Avoidingbusbunchingwithschedulesorwithastate-of-artheadwaycontrolalgorithm

Stopping busbunching!

DriverAssistance• Driversreceivethroughanindustrialtablet,real-timeupdates,messagesfromthecontrolcenter,andautomatizedsuggestiontoincreaseheadwayregularity.

• Driverscanalsomodifyexpeditionparameters,suchasthecommercialstatusoftheexpeditionandthelineassignedtothebus.

• IndustrialtabletisaruggedizeddevicewithAndroidOS.IthasembeddedaGPSreceiverandGSMcommunicationmodule,thereforeitisflexibleenoughtoadapttodifferentneeds.

Ok Hold Go faster Slow down

• All59servicesavailablewiththeheadwaydispatchingmodule

• B22/B14lineswith25industrialtabletsinstalledinbuses

Redbus Project

Results ProjectRedbus

$- $0.10

$0.20

$0.30

$0.40

$0.50

$0.60

$0.70

$- $20,000$40,000$60,000$80,000$100,000$120,000$140,000$160,000$180,000$200,000

Mar-16 Apr-16 May-16 Jun-16 Jul-16 Aug-16 Sep-16

HeadwayRegularityFines

TotalFines Fines/Expedition

• Real-timeupdates&notificationsofbusarrivaltimes,buslocationandbusloads• Multimodalrouteplanner• Userscanevaluatethequalityofservice

Apata:SustainablemobilityAppfortransit+bikeusers

WehaveintegratedBusAssist toanelectronicfarecollectionsystemtoestimatethedemandofeverylineandstop:

• Howmanyusersboardandpay,alightandevadeateachstopatanyperiodoftheday

• Loadprofiles• Givesuggestionstomonitorandsupervisefareevasion

DemandAnalysis

NewChallenges…andOpportunities

• MorecompaniesaroundtheglobeareinterestedonimprovingtheirLOSusingoursolutions

• Howtoimprovetheperceptionofdrivers?

• Howtocommunicatewiththedriver?Imageorvoice?

• Howtopresenttheinformationinthemonitor?

• Gamificationtoimprovecompliance?

– OngoingresearchprojectFONDECYTEngineeringandDesign.

AvoidingBusBunching:FromTheorytoPractice

[email protected]

Department ofTransport EngineeringandLogisticsPontificiaUniversidadCatólicadeChile

Jointworkwith:JuanCarlosMuñozandFelipeDelgado

Avoiding bus bunching: From theory to practice

Education

Transcript of Avoiding bus bunching: From theory to practice