PowerPoint Presentation

Integrated Model of the Urban Continuum: Design, Development and Prototype ImplementationMay 8 12, 2011; Reno, Nevada13th TRB National Transportation Planning Applications ConferenceRam M. Pendyala, Arizona State University, Tempe, AZYi-Chang Chiu & Mark Hickman, University of Arizona, Tucson, AZPaul Waddell, University of California, Berkeley, CABrian Gardner, Federal Highway Administration, Washington DC&The Many Students and Post-Doctoral Researchers Who Make It All Happen

SimTRAVEL: Simulator of Transport, Routes, Activities, Vehicles, Emissions, and Land

BackgroundThree major streams of researchLand use modelingActivity-travel behavior modelingDynamic traffic assignment and simulationCommon thread across innovations in model systemsMicrosimulation approaches involving disaggregate representation of behavioral units, time, and space Modeling urban systems calls for integration of these three streams of researchProgress in integrated modeling slow and devoid of sound behavioral basis (Timmermans, 2003)Ad-hoc statistical coupling and data stitching of disparate model systems

Project DescriptionProject objectiveDevelop a set of methods, computational procedures, data models and structures, and tools for the integration of land use, activity-travel behavior, and dynamic traffic assignment model systems in a microsimulation environment. Universally applicable framework, methods, tools, and data structuresOpen-source enterprise Among first set of projects funded by the FHWA Exploratory Advanced Research Program (EARP)

Design ConsiderationsBehavioralConsistency in behavioral representation, and temporal and spatial fidelityExplicit recognition of inter-relationships across choice processesExample: Response to increase in congestion from home to workShort term - Alter route and/or departure timeMedium term - Adjust work schedule/arrangementsLong term - Change home and/or work locationsComputationalSeparate model systems can take several hours to run a single simulationRun times for integrated model systems could be prohibitiveAdvances in computational power and parallel processing offer hope

Design ConsiderationsDataLand use data available at the parcel levelEmployment and residential data available at the unit-level (e.g., individual employer)Higher-resolution network data with detailed attributes and vehicle classification counts by time-of-dayDetailed activity-travel data including in-home activity informationPolicyHOV/HOT lanes, congestion pricing, parking pricing, fuel price shiftsAlternative work arrangements (flex-hours, telecommuting)Beyond InterfaceMake connections across choice processes within a unified entity (as opposed to loose coupling)

Integrated Model System: LinkagesLand UseTravel DemandTraffic AssignmentLocation choices Activity-travel patternsActivity-travel patterns Network conditionsNetwork conditions Activity-travel patternsNetwork conditions Location choices

Integrated Model System: OverviewFuture Year n + 1Future Year n Base Year Bootstrapping

Base Year SimulationActivity Travel SimulationDynamic Traffic Assignment and SimulationConvergence?TOD OD Trip TimesTOD OD Trip TimesActivity Travel SimulationDynamic Traffic Assignment and SimulationDemand and Supply Convergence?TOD OD Trip TimesActivity Travel SimulationDynamic Traffic Assignment and SimulationDemand and Supply Convergence?TOD OD Trip TimesPopulation Synthesis Land Use ModelPopulation Synthesis Land Use ModelUpdated Travel TimesActivity Travel SimulationDynamic Traffic Assignment and SimulationDemand and Supply Convergence?Population Synthesis Land Use ModelYNYNNYYNPeak/Off-peak Travel TimesPopulation Synthesis Land Use Model

Integrated Model DesignBase Year Simulation: BootstrappingTo obtain time-varying travel skimsStart with peak/off-peak skims from 4-step modelApply model systems sequentially Iteratively run until demand and supply models convergeModel Year Simulation: Integrated ModelLocation choices are simulated once for a model yearActivity-travel patterns are generated and vehicles are routed until both demand and supply side convergence is achievedThe converged time-of-day skims feed into the model simulation for subsequent year This process is repeated on an annual time step

Integrated Model: Supply and Demandt = 1t = 0 mint = 2t = 9Activity-Travel ModelDTA ModelPath is identified and trip is simulatedPerson pursuing activity at destination1440 minutesTrip and Vehicle InformationUpdate O-D Travel Times6 second intervalUpdate Time-Dependent Shortest PathNew Link Travel TimesTrip and Vehicle Information

Supply and Demand Model IntegrationIn each minute, demand model provides a list of vehicle trip records to dynamic traffic assignment modelDynamic traffic assignment model routes and simulates the trips along time-dependent shortest path to destinationsDTA model communicates back arrival times of vehicles that have reached their destinationsSchedules are adjusted and subsequent activity-travel engagement decisions are made based on actual arrival timeThe above steps are repeated to generate activity engagement patterns for all individuals for an entire day

Iterative Process: Feedback LoopsFeedback origin-destination travel times after each iterationMimics learning process of individual from one day to the nextEach iteration represents an adaptation of activity-travel schedule based on past experienceProcess is continued until convergence is achieved both on the demand and supply sideConvergence offers consistency betweenInput travel times used for travel choices in demand modelOutput travel times from network assignment and microsimulation modelHow does one define convergence in the integrated modeling context?

Iterative Process: ConvergenceSupply Side ConvergenceWell-established and incorporated into modeling paradigmsCompare origin-destination travel times from one iteration to the nextUse of relative gap functionDemand Side ConvergenceNo well-established criterion for convergence; simulation results are accepted as stochastic realizations of underlying processCompare daily trip tables at the end of each iteration and compare between iterations to monitor convergenceUse successive averaging schemes as appropriate to bring process to stop

Time-Space Prism ConstraintsHomeWorkUrban SpaceIn-home ActivityTime1vPM Work ActivityIn-home ActivityAM Work ActivityPrism vertices generated by stochastic frontier modelsA prism configured based on the fastest travel mode in the choice setTravel mode availability by time of day and mode continuity checked within and across prisms

Demand Side Convergence: TSP ConstraintsAt more disaggregate level, examine time-space prism vertices for each individual in synthetic populationTime-space prisms are based on origin-destination travel times (travel speeds) and therefore well connected to the supply sideIf time-space prisms show stability from one iteration to the next, process may be approaching convergenceNeed to define notion of stability in a time-space prism context Work in progress

Integrated Model PrototypeSimTRAVEL: Simulator of Transport, Routes, Activities, Vehicles, Emissions, and LandModel SystemsPopGen: Synthetic population generation modelUrbanSim/OPUS: Land use microsimulation model systemOpenAMOS: Activity-based travel microsimulation model systemMALTA: Simulation-based dynamic traffic assignment model FAST-TrIPs: Flexible Assignment and Simulation Tool System for Transit and Intermodal Passengers

SimTRAVEL Design

SimTRAVEL Design: Longer Term Choices

SimTRAVEL Design: Medium Term Choices

SimTRAVEL Design: Short Term Choices

SimTRAVEL Design: Convergence

No

SimTRAVEL SoftwareCompletely open-source and freely available to communityProgramming LanguagesPython used for UrbanSim and OpenAMOSC/C++ used for MALTA/FAST-TrIPsDatabase ManagementPostgreSQL is commonly supported across all model systemsOther database protocols are also supported by individual model systems, including, SQLITE, MySQL, text, HDF5 binary formatGraphic User InterfacesIndividual Model Systems - PyQt4 used for GUIs in PopGen, OpenAMOS and UrbanSim; Java for MALTAIntegrated Model OPUS serves as Master Controller to launch various model systems

SimTRAVEL: Data FlowsLand Use:UrbanSimTravel Demand:OpenAMOSTraffic Assignment:MALTAFAST-TrIPs

Interface: UrbanSim and MALTAData flow between the model systems One way: UrbanSim MALTAData exchangesTravel times and travel costs ()ImplementationMALTA/FAST-TrIPs writes out time-varying O-D skims to text files

Interface: UrbanSim and OpenAMOSData flow between the model systems One way: UrbanSim OpenAMOSData exchangesHousehold location choices ()Fixed activity location choices ()Land use information ()Activity opportunities available at each unitImplementationShared databasesOPUS the software infrastructure used by UrbanSim writes out PostGreSQL database tables at the end of simulation runHousehold and person tables with locations appended and a land use table

Interface: OpenAMOS and MALTAData flow between the model systems Two way: OpenAMOS MALTA/FAST-TrIPsData exchangesMinute-by-minute: Information about trips to be loaded () and information about trips that have arrived at destination ()End of iteration: Travel times and costs ()ImplementationPython embedding (call Python code from C/C++ code)MALTA makes requests for trips at the start of every intervalOpenAMOS generates activity-travel patterns after processing the person arrival information for the intervalMALTA/FAST-TrIPs writes out time-varying O-D skims to text files

Software EnvironmentsSingle workstation environment (implemented)The three model systems run on a single workstationEnables faster and smoother integrationAllows application to small/medium metropolitan regions Distributed computing environment (ongoing work)Various solutions are being exploredRunning the model systems in a cluster computing environment using MPI/OpenMP protocolsAllow application to large/mega metropolitan regions

Test Application: Study AreaMaricopa County, AZ A subarea comprising of City of Chandler, Town of Gilbert and Town of Queen CreekSpatial resolution of analysis is Traffic Analysis Zone (TAZ)The subarea covers a small area505,350 persons residing in 167,738 households159 Traffic Analysis ZonesTest application includesBackground traffic from O-D trip tables + activity-travel demand for synthesized households/persons of three-city regionNon-mandatory activity destinations located throughout Greater Phoenix Metropolitan Region

Study Area

Test ApplicationTo provide efficiency in model testingRuns conducted on five percent population sample Limited to auto mode onlyHourly O-D skims are usedSimulation RunsSequential Application (Bootstrapping procedure)Dynamic ApplicationThe software system is flexible and can accommodate both applications

Sequential ApplicationThis implements the bootstrapping procedureUrbanSim, OpenAMOS, and MALTA are run sequentially and data is passed across modelsArrival information is returned but there is no schedule adjustment done in response to the arrival informationShorter-term activity-travel engagement decisions are not simulated dynamically

Dynamic ApplicationThis implements the proposed integrated model designUrbanSim is run to generate the location choicesOpenAMOS is run to make longer- and medium-term activity-travel engagement decisions, including child dependenciesOpenAMOS-MALTA is run with minute-by-minute handshakingMALTA returns information about all trips that arrived at destinationOpenAMOS processes arrival time to adjust schedules and also to simulate subsequent activity-travel engagement decisions dynamicallyOpenAMOS passes trips that need to be loaded to MALTAAt the end of the iteration, hourly O-D skims are fed back to OpenAMOS for subsequent iteration

Preliminary ResultsFor both sequential and dynamic runs, convergence was achieved after only a few iterationsSequential application: 2 IterationsDynamic application: 3 IterationsConsistent with the low level of demand associated with 5 percent sample and free flow conditions prevail on the network (background traffic not loaded for this experiment)

Preliminary ResultsConvergence Progress: Total Trips Generated

Preliminary ResultsTrip Start Time Distribution

Preliminary ResultsOut-of-home Activity Type Comparison

Computational PerformanceTest runs conducted on a Dell Precision T5400, quad core machine with Intel Xeon Processors and 32 GB RAMProcessing time for 5 percent sample of three-city region with no background traffic (dynamic model)UrbanSim: Few minutesOpenAMOS: 2 hoursMALTA: 2 hoursModel systems scale well and full population runs with background traffic are underway

OpenAMOS: Schedule GenerationA household was randomly chosen to show the formation of schedules in OpenAMOSCharacteristics4 person householdPersons 1 and 4 are adult workersPersons 2 and 3 are non-adult studentsPersons 3 is the only dependent child

OpenAMOS: Schedule GenerationStep 1: Fixed Activity Schedule

OpenAMOS: Schedule GenerationStep 2: Reconcile Fixed Activity Schedule

OpenAMOS: Schedule GenerationStep 3: Schedule Adjustment to Reflect Daily Status and Dependent Child Activities

OpenAMOS: Schedule GenerationStep 4: Adjust Schedules to Introduce Appropriate Travel Episodes

OpenAMOS: Schedule GenerationStep 5: Generate Flexible Activities and Allocate Dependent Child Activities to Adults

OpenAMOS: Schedule GenerationStep 6: Adjust Schedules to Fill the Entire Day

OpenAMOS: Schedule GenerationStep 7: Final Activity and Trip Schedule

Scenario AnalysisModel testing involves evaluation of of policies for the full populationScenario typeConditionsSocio-economic scenariosChange in population across the regionChange in population in some parts of the regionChange in employment densityHighway scenariosChange in fuel price or value pricing schemeChange in link capacity (Example: link removal)Transit scenariosChange in transit faresChange in transit service frequencyTravel demand management scenariosAlternative work schedulesTelecommutingIntroducing HOV lanesIntroducing HOT lanes

PopGen Software

UrbanSim Software

UrbanSim Software

UrbanSim Software

OpenAMOS Software

OpenAMOS Software

OpenAMOS Software

OpenAMOS Software

OpenAMOS Software

Google Earth Visualization

MALTA Software

MALTA Software

Wiki: simtravel.wikispaces.asu.eduCode RepositoriesPopGen: http://code.google.com/p/populationsynthesis/OpenAMOS: http://code.google.com/p/simtravel/UrbanSim: https://svn.urbansim.org/MALTA: https://dev.urbansim.org/malta

Resources

Ongoing WorkCalibration and validation of individual model systems as well as integrated systemRun the integrated model for the full population including background trafficAddress computational issues Explore distributed environmentsScenario analysis

AcknowledgementsProject sponsorFederal Highway Administration, Exploratory Advanced Research ProgramProject Manager: Brian GardnerPeer Review PanelKonstadinos Goulias, David Boyce, Maren Outwater, John Gliebe, Vladimir Livshits, Keith Killough, Aichong Sun

Questions?Work/School/Pre-school Location Choices

Synthetic Population Generator

Vehicle Ownership Model

Fixed Activity Prism Generator

Child Daily Status and Allocation Model

Activity Skeleton Reconciliation

Activity-Travel Simulator including Activity-Travel Dimensions Vehicle Choice Model Activity Accompaniment Model (Solo Vs Joint)

Activity-travel Pattern Reconciliation

Simulated for the whole day

Simulated for each time slice (n minutes) within a day

Is convergence achieved?Check Time-use/ Check OD/TT tables aggregated to 30 min during Peak and Off-peak

Time-use Utility

Start simulation for Base Year

No

After simulation is run for the whole day

End simulation for Base Year

Yes

Adult Daily Status

Longer term choices simulated for the base year

MALTA: Skims for the whole day from previous years run

UrbanSim: Location Choices

MALTA: Arrival times

MALTA: Travel Information

Simulate on Network

MALTA: Skims for the whole day from previous iteration (day)

Work/School/Pre-school Location Choices

Synthetic Population Generator

Vehicle Ownership Model

Start simulation for Base Year

Longer term choices simulated for the base year

MALTA: Skims for the whole day from previous years run

UrbanSim: Location Choices

Fixed Activity Prism Generator

Child Daily Status and Allocation Model

Activity Skeleton Reconciliation

Simulated for the whole day

Adult Daily Status

MALTA: Skims for the whole day from previous iteration (day)

Activity-Travel Simulator including Activity-Travel Dimensions Vehicle Choice Model Activity Accompaniment Model (Solo Vs Joint)

Activity-travel Pattern Reconciliation

Simulated for each time slice (n minutes) within a day

MALTA: Arrival times

MALTA: Travel Information

Simulate on Network

Is convergence achieved?Check PrismsCheck OD/Skim tables

Time-use Utility

After simulation is run for the whole day

End simulation for Base Year

Yes