Data Driven Aircraft Trajectory Prediction...
Transcript of Data Driven Aircraft Trajectory Prediction...
DARTDataDrivenAircraftTrajectoryPredictionResearchNEWSLETTERNo.2-November2017
DARTATAGLANCECall SESAR-2-2015Objective DataScienceinATM
Duration June2016-June2018
Contact
ProjectCoordinatorProf.GeorgeA.VourosUniversityofPiraeusResearchCenter,[email protected]
ScientificManagerDrGeorgFucksIAISFraunhofer,[email protected]
URL: http://www.dart-research.euSHAPINGTHEFUTURE:DARTVISIONDART explores the applicability of a collection of machinelearning and agent- basedmodels and algorithms to derive adata-driven trajectory prediction capability. Advanced visualanalytics techniques are used to facilitate data exploration,quality assessment, and algorithms parameters and featuresselection.Duringthe1stconsultationmeetingwehadwithstakeholdersonSeptember,weshapedtheDARTvision:Toadvancecollaborativedecision-making processes that support multi-objectiveoptimizationtakingtherequirementsofthedifferentactors intheATMsystemintoaccountattheplanningphase(i.e.fewdaysbeforeoperation):•Aircraft Operators (AOs): Minimizing the cost thoughtmaximizingtheadherencetotheairlinespreferredFPs.•Network Manager (NM): Decide which flights to modify toresolvesectorimbalancesandpotentialconflicts.•AirNavigationServiceProviders(ANSPs):Minimizingthesectorimbalancesandpotentialconflicts.Theoverallworkflowincollaborativedecisionmakingisshowninthefollowingpicture.
DARTDATASOURCES(UPDATED)
DART
StrategicDelayCosts
SurveillanceDataIFSbyEnAire
WeatherData:NOAAforecasts,SIGMET,TAF
FlightPlansDDR-2Spanish
Operational Data
AirspaceStructure
DDR-2SpanishOperational
Data
ReconstructedTrajectoriesBR&T-E
TrajectoryPredictorusingFlightPlans
Aircraft IntentDescriptionsBR&T-E
CHALLENGEDARTaimstodeliverdata-driventechniquestoimprovetheperformanceandaccuracyofsingleandmultipletrajectorypredictions,accountingforATMnetworkcomplexityeffects.
RESEARCHISSUES•Whatarethesupportingdatarequiredforrobustandreliabletrajectorypredictions?•Whatisthepotentialofdata-drivenmachinelearningalgorithmstosupporthigh-fidelityaircrafttrajectoryprediction?•HowthecomplexnatureoftheATMsystemimpactsthetrajectorypredictions?•HowcanthisinsightbeusedtooptimizetheATMsystem(e.g.adherencetopreferredflightplans,betterutilizationof
CTPCollaborativeTrajectoryPrediction
Imbalancesandconflicts
detection(ANSP)
AOsp
referr
edFP
s
IterativeOptimizationprocess
Selecttheflightstomodify(NM)
STPSingleTrajectory
Prediction
Trajectorypredictionsfromanindividual
trajectoryperspective
STPpredictnewsingletrajectoriesfortheseflights
CTPdetectionofimbalancesandconflicts,selectionof
flighttomodifyandpropositionofnewFPs
Optim
altra
jector
ies
AOsproposenewFPfor
theseflights
DARTALGORITHMSHIDDENMARKOVMODELS(HMM):trajectorypredictionusingdiscretespatio-temporalstate-transitionmodels.SIMILARITY-BASEDRETRIEVAL(CLUSTERING):trajectorypredictionbymeansofanN-dimensionalclusteringtaskofdataseries.
REGRESSIONMODELS:trajectorypredictionusinganN-dimensionalregressiontaskw.r.t.dataseries.REINFORCEMENTLEARNING:MarkovDecisionProcess(MDP/POMDP):trajectorypredictionusingastate-action-rewardmodelinordertoobtainanoptimalpolicyonanyinstantoftheflight.JOINT(COLLABORATIVE)TRAJECTORYPREDICTIONS:Formulatetheproblemasamulti-agentMarkov-Decision-Processandsolveitviacollaborative,decentralizedreinforcementlearningmethods.
COMMUNICATION&DISSEMINATIONACTIVITIES
• DARTPresentedinSID2017,Belgrade,Servia,thearticle
DART:AMachine-LearningApproachtoTrajectoryPredictionandDemand-CapacityBalancing
• AlongsidetheInternationalConferenceforResearchinAir
Transportation(ICRAT)2018,DART(incollaborationwithdatAcronproject)willbeorganizingaworkshoponData-EnhancedTrajectoryBasedOperations.Pleasevisitthefollowingwebpageforadditionaldetails:http://icrat.org/icrat/upcoming-conference/data-tbo-workshop/
VISUALIZATIONSANDVISUALANALYTICSVisualization, interaction techniques and interfacessupport the exploration and evaluation of results oftrajectory prediction algorithms, particularly,comparisonofpredicted trajectories to realonesandcomparison of predictions obtained with differentalgorithmsordifferentparametersettings.Visuallysupporteddetectionofclustersoftrajectoriesandflightplans:
Historical data provide recurrent patterns oftrajectories (enriched with contextual information –e.g.weather)thatdata-drivenmethodslearn.Predictionofasingletrajectory involveschoosingthepattern thatfitsbetteraflightplan w.r.t.contextualdata. Visual explorationof results is beneficial for analystsandstakeholders(e.g.AircraftOperators),tofine-tuneprediction algorithms and to understand better thereasonsfordeviations.
Linkingpointstocomparetrajectories
Theabovehistogramshowsthestatisticaldistributionoftheproportionsofthematchedpointsbetween
trajectories(e.g.proportionsarehigherforthebluetrajectorythanfortheredone).
RESULTS&BENEFITS•Data-driventrajectorypredictioncapabilities;
•Agent-basedmultipletrajectorypredictionabilities;
•Interactivevisualinterfacesforsupportinginteractiveexplorationofmodellingresultsinspaceandtime,supportingdecisionmaking.
This project has received funding from the SESAR Joint Undertaking under grant agreement No 699299 under European Union’s Horizon 2020 research and innovation programme
The opinions expressed herein reflect the author’s view only. Under no circumstances shall the SESAR Joint Undertaking be responsible for any use that may be made of the information contained herein.