Travel demand models: from trip- to activity-based - PowerPoint Presentation

1. Travel demand models: from trip- to activity-based9 december 2019Guest Lecture by Luuk BrederodeFor: course CIE5802-09 | Advanced transportation modelling, Delft University

2. ContentsIntroduction: travel demand modelsWhy tour- or activity-based? Tour/Activity based (BRUTUS) vs trip based (traditional 4-step)Methodology of BRUTUS in more detailbreakPro’s and cons of activity based in practiceChallenges ahead / vacancy internship assignments2

3. IntroductionTravel demand model types

4. Travel demand(OD-matrices)Infrastructural supply(Network)Demand ModelsSupply ModelsAssignmentLink flows, link speeds, route travel times, congestion patterns, emissions, etc.Demand and supply models4

5. Change in behavior is triggered by change of circumstances e.g.: travel time on a route decreases; train ticket prices increase,…Each change has its own response time:One does not change his mode or destination overnight (mainly due to mode and activity availability and habit)Changing origin mostly involves relocation…5SecondsMinutesHoursDaysWeeksMonthsYearsDecadesResponse time to changed circumstancesChange of OriginChange of destinationChange of modeChange of departure timeChange of routeDemandModelsLand-Use ModelsAssignment ModelsDemand and supply models

6. 6Demand and supply modelsDemand Model applicationsSupply Model applicationsOperationalTacticStrategicStategicTacticOperationalThe application context determines the importance of the different model components!

7. Activity BasedActivity schedules: persons to zoneTrip BasedTour BasedTypes of demand modelsUnit of demandTrips: zone to zoneTours: segment per zone to zone

8. Aggregated (macroscopic):Inhabitants in a zone all exhibit the same ‘average behavior’Gravity modellingDisaggregated (mesoscopic):Different (person/household-) segments exhibit different behaviorRandom Utility MaximizationMicrosimulation (microscopic):Different persons/households exhibit different behaviorRandom Utility Maximization / Decision trees / Machine learningTypes of demand models8Behavioural aggregation and modelling paradigms

9. 9Types of demand modelsConsidered constraintsTrip BasedActivity BasedTour BasedTrip endTour consistencySpace/TimeHouseHoldParking***Vovsha (2019) argues that parking constraints cannot be modelled by these approaches

10. Trip BasedActivity BasedTour Based10Types of demand modelsTrip GenerationDestination ChoiceTOD choiceMode choiceTour Generation1th Destin. ChoiceTOD choice (inb/outb)Mode choice2nd(+)Destin. Choices1th Destin. ChoiceTOD choice (inb/outb)ModeGroup choice (tour)2nd(+)Destin. ChoicesSchedule GenerationMode choice (trip)Departure time choice*Flow charts interpreted from Vovsha (2019)Integration/sequencing of choices

11. 11Travel representationUsed in software (NL)Trip-based modelssingle OD travelOmniTRANS, INDY, MarpleTour-based models(multi-destination) round tripLMS/NRM, Omni- TRANS (Brutus)Activity-based modelsmultiple (multi-destination) round trips over one or more daysAlbatross, MatsimTypes of demand modelsImplementations used in NL

12. Travel demand (OD-matrices)AM peakfraction / modelAssignment Off-peakAssignment AM peak12Temporal aggregationDemand and supply modelsInfrastructural supply(Network)Demand ModelsSupply ModelsAssignment AM peakLink flows, link speeds, route travel times, congestion patterns, emissions, etc.Day-level demand modelAM peakfraction / modelAM peakfraction / modelHousehold constraintsConsistency constraintsSpace/time constraintsTrip-End constraintsParking constraints

13. ConcludingUnit of demand (trip, tour or activity schedule) is the most important property of demand models, but just one of them. There’s alsoBehavioral aggregation appliedModelling paradigm(s) usedConstraints (and their temporal aggregation) consideredIntegration/sequencing of behavioral choices usedAnd you can mix them all too….13Types of demand models

14. Why tour- or activity based?

15. It adds real world constraints, eliminating weird phenomena that can occur in trip-based modelsAbandoning your car at a destination, using public transport for the ride back Why tour- or activity based???Simultaneous use of a single vehicle for multiple trips by different household membersAlways returning home in between activities15

16. When using more segments and/or more sparsly populated zones, activity based might be computationally more efficient:Why tour- or activity based?DisaggregatedActivity basedReference situation: More agents than segments disaggregated computationally favorableIncreased number of segments (e.g. to be able to distinguish electric car owners)is disaggregated still computationally favorable?(additionally) spatially refined zoning systemActivity based computationally favorable16

17. Why would you have many segments?Because the number of segments equals:where : set of segmentvariables (e.g.: [age, gender]) : set of classes for variable (e.g. [<18, >=18], [male/female])This means that adding segmentvariable yields times more segments. For instance: LMS/NRM uses 311 different person- and household segments based on 6 variables:  Why tour- or activity based?17

18. Tour consistencySpace/Time constraintsHousehold constraintsAvailability of an autonomous vehicle for a trip is dependent on location of person, vehicle and usage by other people in the householdPublic transport (or MaaS) may be used depending on mode choices of another household members. When working from home, the daily trip schedule changesAvailability and price of demand responsive transport will depend on demand from other travelers.18Why tour- or activity based?It better fits today’s questions on transport and mobility

19. Brutus vs 4-stage model Methodological comparison

20. 2009: 1st version of BRUTUS created by Strafica (now: Ramboll)Tour based*, pragmatic demand model in RSamples where possible, models where neededElegantly constructs trip chainsElegantly handles dependencies between mode and destination choice Modular structure allows for future improvements2018: Goudappel Group joins development on BRUTUS to:Make BRUTUS more applicable in strategic context (stability, robustness)Make BRUTUS more scalableCreate first application in Hybrid form (with aggregated gravity model) for municipality of Almere20BRUTUS in a nutshell*it also has some activity based features: it is a microsim and it includes household constraints

21. Model ApplicationPopulation synthesizerTrip chain samplerSimulator(assignment)Model EstimationDestination choice modelMode choice modelModel ApplicationTrip generation modelGravity model(assignment)Model EstimationTrip generation modelMode/destination choice modelBRUTUSAggregated 4-stage modelModelling Stages21

22. Zoning systemsBRUTUS: 200*200m grid4Stage: ‘regular’ zoning22

23. BRUTUSPopulation SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeTrip generationProductions / AttractionsGravity ModelTrip table per modeAggregated 4-stage modelComparison: application23Trip GenerationDestination ChoiceTOD slicingMode choiceDestination ChoiceTOD slicingMode choiceSchedule Sampling

24. Brutus Methodology

25. Synthetic households per 200x200m grid celPopulation synthesizer1Zonal targets per 200x200m grid cel (person level)Employed (yes/no)Student (yes/no)Age group (5 classes)2Zonal targets per 200x200m grid cel (household level)Househ. size (6 classes)Number of cars in household (0-4+)Zonal targets1Distribution of 15 person segments (from OViN)Zonal targets1Synthetic persons per 200x200m grid celPopulation SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per mode25Iterative Proportional fittingZonal targets2Distribution of 24 household person segments (from OViN)Zonal targets2Iterative Proportional fittingDutch Mobility Panel microdata containing household compositionsSynthetic populationper 200x200mgrid celIterative Proportional updating and MILP Solver

26. Trip chain samplerPopulation SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeFor each person in the synthetic populationRandomly draw a person in the same segment from OViNIn preparation for the destination choice model: copy for each trip in the trip chain:the activity on the destination, the order number of the activity; and the activity durationIn preparation for the mode choice model: copy some variables that are not defined by the segment:Drivers license possession, Gender; and income26

27. rnidHhidpidageagegemployedstudentfam_sizecar_owndrivlincomemalechaintripitimeitypejtypestaytimeifixedjfixed27111101FALSETRUE62FALSENATRUE118.515405TRUEFALSE27111101FALSETRUE62FALSENATRUE1215.5851745FALSETRUE27111201FALSETRUE62FALSENAFALSE118.2515258TRUEFALSE27111201FALSETRUE62FALSENAFALSE1212.6351922FALSETRUE27111311FALSEFALSE62FALSENAFALSE1110.51715TRUEFALSE27111311FALSEFALSE62FALSENAFALSE1210.9171195FALSETRUE27111311FALSEFALSE62FALSENAFALSE2314.331415TRUEFALSE27111311FALSEFALSE62FALSENAFALSE2414.7541795FALSETRUE27111411FALSEFALSE62FALSENAFALSE118.251410TRUEFALSE27111411FALSEFALSE62FALSENAFALSE128.58341360FALSETRUE27111411FALSEFALSE62FALSENAFALSE2314.751410TRUEFALSE27111411FALSEFALSE62FALSENAFALSE2415.0841775FALSETRUE27112521FALSETRUE41FALSENATRUE118.58315290TRUEFALSE27112521FALSETRUE41FALSENATRUE1213.5512135FALSEFALSE27112521FALSETRUE41FALSENATRUE1316121720FALSETRUE27112631FALSETRUE41FALSENATRUE118.33315215TRUEFALSE27112631FALSETRUE41FALSENATRUE121251960FALSETRUE27112741FALSEFALSE41FALSENATRUE118.16715400TRUEFALSE27112741FALSEFALSE41FALSENATRUE121551780FALSETRUEFrom population synthesizerFrom trip chain samplerTrip chain constraintsPopulation SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per mode27

28. Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeAlgorithm for destination choiceInitialize all non-home activities to not-fixed. While there still are non-fixed activities in the chain:Determine activity type i: the next activity with longest duration that has not been fixed yetDetermine last fixed activity type h before, and first fixed activity type j after the current activity i. Triplet (h,i,j) defines which destination choice model (parameters) are used to determine location of i, given h and j using the destination choice model.Trip simulator28

29. Determines the probability of choosing trip end point i, given previous fixed end point h and next fixed end point j: Where utility is defined as: With : travel time from h to i : travel time from I to j : land use at trip end point : parameter to be estimated for each segment () Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeTrip simulatorDestination choice model29

30. Trip chain constraintspagina 30HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:001112Workhome-1211Homedrop off/pick up00:151212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-AlgorithmInput: The trip chain sampler provides trip chain constraints: home location, activity types, activity order and activity durations for all trip chains made by the synthetic population.Destination choice model results

31. Trip chain constraintspagina 31HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00True1112Workhome-True1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-TrueAlgorithmInitialization: Set home activities to fixedDestination choice model results

32. Trip chain constraintspagina 32HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00True1112Workhome-True1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-TrueAlgorithmSelect next chain or next person and its home locationHomeWork8:00Destination choice model results

33. Trip chain constraintspagina 33HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00True1112Workhome-True1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-TrueAlgorithmWhile there still are non-fixed activities:Determine activity type i: the longest unfixed activity in the chainWork8:00HomeDestination choice model results

34. Trip chain constraintspagina 34HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00True1112Workhome-True1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-TrueAlgorithmWhile there still are non-fixed activities:Determine triplet (h,i,j) where h is the last fixed activity type before i; and; j is first fixed activity type j after i.Work8:00HomeDestination choice model results

35. Trip chain constraintspagina 35HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00True1112Workhome-True1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-TrueAlgorithmWhile there still are non-fixed activities:Apply destination choice model using parameters for {home,work,home} to choose location of i, considering:travel time from h to candidate i, travel time from candidate i to j Number of jobs at candidate ISet location i to fixedWork8:00HomeDestination choice model results

36. Trip chain constraintspagina 36HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upShopping00:201215shoppinghome-TrueAlgorithmWhile there still are non-fixed activities:Apply destination choice model using parameters for {home,work,home} to choose location of i, considering:travel time from h to candidate i, travel time from candidate i to j Number of jobs at candidate ISet location i to fixedWork8:00HomeDestination choice model results

37. Trip chain constraints7:00pagina 37HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upShopping00:201215shoppinghome-TrueAlgorithmSelect next chain or next person and its home locationWorkDrop offPick upShopping00:1500:0500:20HomeDestination choice model results

38. Trip chain constraintspagina 38HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-TrueAlgorithmWhile there still are non-fixed activities:Determine activity type i: the longest unfixed activity in the chain7:00WorkDrop offPick upShopping00:1500:0500:20HomeDestination choice model results

39. Trip chain constraintsWorkpagina 39HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:001213workdrop off/pick up00:051214drop off/pick upshopping00:201215shoppinghome-True7:00Drop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine triplet (h,i,j) where h is the last fixed activity type before i; and; j is first fixed activity type after i.HomeDestination choice model results

40. Trip chain constraintsWorkpagina 40HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:00True1213workdrop off/pick up00:05True1214drop off/pick upshopping00:201215shoppinghome-True7:00Drop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Apply destination choice model using parameters for {home, work, home} to choose location of i, considering:travel time from h to candidate i, travel time from candidate i to j Number of jobs at candidate ISet location i to fixedHomeDestination choice model results

41. Trip chain constraintsWorkpagina 41HhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:00True1213workdrop off/pick up00:05True1214drop off/pick upshopping00:201215shoppinghome-True7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine activity type i: the longest unfixed activity in the chainDestination choice model results

42. pagina 42Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:00True1213workdrop off/pick up00:05True1214drop off/pick upshopping00:201215shoppinghome-True7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine triplet (h,i,j) where h is the last fixed activity type before i; and; j is first fixed activity type after i.Destination choice model results

43. pagina 43Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:00True1213workdrop off/pick up00:05True1214drop off/pick upshopping00:20True1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Apply destination choice model using parameters for {work, shopping, home} to choose location of i, considering:travel time from h to candidate i, travel time from candidate i to j Shopping floor size at location iSet location i to fixedDestination choice model results

44. pagina 44Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:00True1213workdrop off/pick up00:05True1214drop off/pick upshopping00:20True1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine activity type i: the longest unfixed activity in the chainDestination choice model results

45. pagina 45Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15True1212drop off/pick upwork07:00True1213workdrop off/pick up00:05True1214drop off/pick upshopping00:20True1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine triplet (h,i,j) where h is the last fixed activity type before i; and; j is first fixed activity type after i.Destination choice model results

46. pagina 46Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15TrueTrue1212drop off/pick upwork07:00TrueTrue1213workdrop off/pick up00:05True1214drop off/pick upshopping00:20True1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Apply destination choice model using parameters for {home, dropOff, work} to choose location of i, considering:travel time from h to candidate i, travel time from candidate i to j Shopping floor size at location iSet location i to fixedDestination choice model results

47. pagina 47Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15TrueTrue1212drop off/pick upwork07:00TrueTrue1213workdrop off/pick up00:05True1214drop off/pick upshopping00:20True1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine activity type i: the longest unfixed activity in the chainDestination choice model results

48. pagina 48Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15TrueTrue1212drop off/pick upwork07:00TrueTrue1213workdrop off/pick up00:05True1214drop off/pick upshopping00:20True1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Determine triplet (h,i,j) where h is the last fixed activity type before i; and; j is first fixed activity type after i.Destination choice model results

49. pagina 49Trip chain constraintsWorkHhidpidchaintripitypejtypeDurationi_fixedj_fixed1111Homework08:00TrueTrue1112Workhome-TrueTrue1211Homedrop off/pick up00:15TrueTrue1212drop off/pick upwork07:00TrueTrue1213workdrop off/pick up00:05TrueTrue1214drop off/pick upshopping00:20TrueTrue1215shoppinghome-TrueTrue7:00HomeDrop offPick upShopping00:1500:0500:20AlgorithmWhile there still are non-fixed activities:Apply destination choice model using parameters for {work, pick up, shopping} to choose location of i, considering:travel time from h to candidate i, travel time from candidate i to j Shopping floor size at location iSet location i to fixedDestination choice model results

50. Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeAlgorithm for mode choiceThe destination choice model is run separately for each modegroup:In Almere model: Pedestrian, bike, transit, car driver, car passengerPer modegroup allowed combinations of modes for each trip-tuple can be specified to include mode-chain constraints. E.g. for modegroup transit:Trip simulatormodegroupmode(h,i)mode(i,j)transitwalkpublic transporttransitpublic transportwalktransitpublic transportpublic transport50

51. 51Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeAlgorithm for mode choiceThis yields optimum chains per modegroup (different destinations!!)The mode choice model is applied to determine the chosen modegroup

52. Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeDetermines the probability of choosing mode m, given a tripchain c: Where utility is defined as: With : total travel time when using mode m for trip chain c : explanatory variables (traveltime ratios, land use, car ownership) : average destination choice logsum of trips in chain c, normalized to max land use M, calculated as: : parameter to be estimated for each explanatory variable and each mode Trip simulatorMode choice model52

53. Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modeOutputrnidHhidpidageagegemployedstudentfam_sizecar_owndrivlincomemalechaintripitimeitypejtypestaytimeRnid_iRnid_jmode27111101FALSETRUE62FALSENATRUE118.51540559001821Car27111101FALSETRUE62FALSENATRUE1215.585174547939907Car27111201FALSETRUE62FALSENAFALSE118.251525842336347Car27111201FALSETRUE62FALSENAFALSE1212.635192276755846Car27111311FALSEFALSE62FALSENAFALSE1110.5171586297716Car27111311FALSEFALSE62FALSENAFALSE1210.917119578427856Car27111311FALSEFALSE62FALSENAFALSE2314.33141553103798Car27111311FALSEFALSE62FALSENAFALSE2414.75417954315666Car27111411FALSEFALSE62FALSENAFALSE118.25141028129563Car27111411FALSEFALSE62FALSENAFALSE128.5834136021937323Car27111411FALSEFALSE62FALSENAFALSE2314.75141038752146Car27111411FALSEFALSE62FALSENAFALSE2415.08417752128120Car27112521FALSETRUE41FALSENATRUE118.5831529034224083Car27112521FALSETRUE41FALSENATRUE1213.551213522398943Car27112521FALSETRUE41FALSENATRUE131612172020311482Car27112631FALSETRUE41FALSENATRUE118.3331521548574137Car27112631FALSETRUE41FALSENATRUE12125196080694704Car27112741FALSEFALSE41FALSENATRUE118.1671540091864092Car27112741FALSEFALSE41FALSENATRUE12155178019137947CarFrom population synthesizerFrom trip chain samplerFrom trip simulator53

54. break54

55. Pro’s and con’straditional vs tour/activity based

56. BRUTUS enables highly detailed analysis of the socio-economic effects of policies; We can ”interview” a section of network and ask people where they live and work, income, etc.Population SynthesizerSynthetic populationTripChain SamplerTrip chain constraintsTrip SimulatorTrip table per modePro: segmentation available56

57. We can now model scenarios that we couldn’t before: Population composition. E.g. effects of: Aging populationDecreasing household sizesTypical behavior of millennials/generationX/generationWhateverPotential for new mobility concepts. E.g.: potential for: Mobility hubs MaaS subscriptionsEffect of Connected Autonomous Vehicles on mode/destination choice…57Pro: added policy variables

58. Yes, (much) more realism due to constraints and spatio/temporal detail, but:Discrete hence stochastic resultsNot possible to enforce solution conditions such as max entropyMuch more (survey) data required for estimation58Cons

59. Spread in old BRUTUS population synthesizer (current IPF-based version is deterministic, hence no spread)59Con: stochastic resultsNumber of synthetic persons in a segment in a zone (mean of 2000 pop synth replications)Number of synthetic persons in a segment in a zone (5th percentile pop synth replication for that segment/zone)Conclusion (rule of thumb): the calculated number of people of a certain segment in a certain zone can be upto 51% lower than the average from 2000 replications

60. Spread in old BRUTUS population synthesizer (current IPF-based version is deterministic, hence no spread)60Number of synthetic persons in a segment in a zone (mean of 2000 pop synth replications)Number of synthetic persons in a segment in a zone (95th percentile pop synth replication for that segment/zone)Conclusion (rule of thumb): the calculated number of people of a certain segment in a certain zone can be upto 56% higher than the average from 2000 replicationsCon: stochastic results

61. 61Con: more data required

62. Challenges aheadStudents wanted!

63. Developments of past yearReplaced sequential population synthesizer with IPF/IPU based versionCreated a prototype of a tour based gravity model based on formulation of (Honma 2010)Work-in-progress: replace trip-chain sampler with trip chain generator Work-in-progress: re-estimate mode and destination choice models with NVP data and better segmentation.Also include drivers license possession, gender and income63

64. Extending applicability of an agent-based destination choice modelExperiences with BRUTUS have shown that the destination choices are replicated well, but only within a limited range of trip lengths. Therefore, for application, the agent-based model is currently coupled with an aggregated model which describes the long trips. The goal of this research is extend BRUTUS’ applicability, such that all trip lengths can be sufficiently captured. It seems natural to do this by transformations of the variables within the utility function and / or extension of the utility function with variables that better capture the destination choice sensitivities on different trip lengths, but other methods should be looked in to as well. https://www.dat.nl/vacatures/afstudeeropdrachten/ Assignment 1/364

65. Adding attraction constraints to an agent-based travel demand modelhttps://www.dat.nl/vacatures/afstudeeropdrachten/ Although behaviorally more accurate, BRUTUS does not contain trip attraction constraints. Although the trip production constraints will generally keep the attractions in balance, the number of trips arriving at a destination might not match the number of arrivals that is to be expected given the number of jobs, shops and other attraction variables. The goal of this research is to add trip attraction constraints to the current BRUTUS implementation in such a way that reproducibility (or even: uniqueness) of the model outcomes, scalability and calculation time are unaffected or improved. Assignment 2/365

66. Removing stochasticity from an agent-based travel demand modelIn strategic applications transport model outcomes are used by the model analyst who compares different future scenario’s to a reference situation and translates insights from this comparison into advice to decision makers. When the model contains stochasticity, differences may be caused by differences in the inputs (which is desirable) or due to stochastic randomness (which is not desirable). The goal of this research is to isolate the stochastic processes within the current BRUTUS trip-yield implementation and replace it with a trip-type-choice model that yields stable results. https://www.dat.nl/vacatures/afstudeeropdrachten/ Assignment 3/366

67. Questions? More info on the assignments?Luuk Brederodelbrederode@dat.nl+31 627369830

68. LiteratureVovsha, P., 2019. Decision-Making Process Underlying Travel Behavior and Its Incorporation in Applied Travel Models, in: Bucciarelli, E., Chen, S.-H., Corchado, J.M. (Eds.), Decision Economics. Designs, Models, and Techniques for Boundedly Rational Decisions. Springer International Publishing, Cham, pp. 36–48. https://doi.org/10.1007/978-3-319-99698-1_5Honma, Y., Kurita, O., Taguchi, A., 2010. SPATIAL INTERACTION MODEL FOR TRIP-CHAINING BEHAVIOR BASED ON ENTROPY MAXIMIZING METHOD. Journal of the Operations Research Society of Japan 53, 235–254. https://doi.org/10.15807/jorsj.53.235pagina 68