Presentation on theme: "Andrea D’Ariano, ROMA TRE University"— Presentation transcript:
1Andrea D’Ariano, ROMA TRE University Rolling Horizon Approach For Aircraft Scheduling In The Terminal Control Area Of Busy AirportsAndrea D’Ariano, ROMA TRE UniversityISTTT, 22/03/20171
2Modeling a Terminal Control Area Solution Framework and Algorithms Presentation outlineIntroductionModeling a Terminal Control AreaSolution Framework and AlgorithmsComputational ExperimentsConclusions and Ongoing ResearchThis work was partially supported by the Italian Ministry of Research, project FIRB “Advanced tracking system in intermodal freight transportation”.
3Air Traffic Control (ATC) Air traffic control must ensure safe, ordered and rapid transit of aircraft on the ground and in the air segments.With the increase in air traffic [*], aviation authorities are seeking methods (i) to better use the existing airport infrastructure,and (ii) to better manage aircraftmovements in the vicinity of airports during operations.Indicating Standard Itinerary (IFR) = num passengersSFT: Short-Term Forecast (September 2009)Tween towers, usa economy crisis[*] Source: EUROCONTROLShort-term forecast 2009
4Status of the current ATC practise Airports are becoming a major bottleneck in ATC operations.The optimization of take-off/landing operations is a key factor to improve the performance of the entire ATC system.ATC operations are still mainly performed by human controllers whose computer support is most often limited to a graphical representation of the current aircraft position and speed.Intelligent decision support is under investigation in order to reduce the controller workload (see e.g. recent ATM Seminars).
5Literature: Aircraft Scheduling Problem (ASP) Chris Potts et al. 2011Detailed(e.g. Bianco, Dell’Olmo, Giordani)Basic(e.g. Bertsimas, Lulli, Odoni )Dynamic(e.g. Beasley, Ernst)Static(e.g. Dear, Hu, Chen)Existing Approaches
6Literature: Research needs & directions Aircraft Scheduling Problem (ASP) in Terminal Control Areas:Most aircraft scheduling models in literature represent the TCA as a single resource, typically the runway. These models are not realistic since the other TCA resources are ignored.We present the “alternative graph” approach for the accurate modelling of air traffic flows at multiple runways and airways.This approach has already been applied successully to control railway traffic for metro lines and railway networks.
7Our approach for TCAs Design, implementation and testing of: a dynamic (rolling horizon) settinga detailed (alternative graph) modelingheuristic and exact (branch & bound) ASP algorithmsResearch questions:how does the traffic control system react when disturbances arise,when and how is it more convenient to reschedule aircraft in the TCA,which algorithm performs best in terms of delay and travel time minimization,which algorithm is the less sensitive to disturbances.
8Modeling a Terminal Control Area Solution Framework and Algorithms Presentation outlineIntroductionModeling a Terminal Control AreaSolution Framework and AlgorithmsComputational ExperimentsConclusions and Ongoing ResearchThis work was partially supported by the Italian Ministry of Research, project FIRB “Advanced tracking system in intermodal freight transportation”.
10The Alternative Graph (AG) Model Mascis & Pacciarelli 2002The quality of a schedule is measured in terms of maximum delay minimization (limiting the delay caused by potential conflicts).Fixed constraints in F model feasible timing for each aircraft on its specific route, plus constraints on each resource of its route.Alternative constraints in A represent the aircraft ordering decision at air segments and runways, plus decisions on holding circles.A feasible schedule is an event graph with no positive length cycles.Conflict-free schedule = no positive length cycles and all potential conflicts are solved (alternative arcs are selected)
11AG Model Release date αA (αA = expected entry time of aircraft A) A1 SegmentsAG ModelHolding CirclesCommonGlide PathRunwaysA1TOR410131659RWY 35LMBR2611157143SRN17812RWY 35RA1*αADevi dire che per avere un approccio microscopico proponiamo un modello basato sull’alternative graph,Si tratta di un modello più dettagliato di quello di Bianco ma cmq basato sul job shop schedulingRelease date αA(αA = expected entry time of aircraft A)
12AG Model Entry due date βA ( βA = - αA ) A1 Holding Circles Air SegmentsCommonGlide PathRunwaysA1TOR410131659RWY 35LMBR2611157143SRN17812RWY 35RA1*αAβAEntry due date βA( βA = - αA )
13AG Model (A4, A1) No holding circle (holding time = 0) (A1, A4) Holding CirclesAirSegmentsCommonGlide PathRunwaysA1TOR410131659RWY 35LMBR2611157143SRN17812A1A4*αAβAδRWY 35R-δ(A4, A1)No holding circle (holding time = 0)(A1, A4)Yes holding circle (holding time = δ)
14AG Model Time window for the travel time in each air segment Holding CirclesAirSegmentsCommonGlide PathRunwaysA1TOR410131659RWY 35LMBR2611157143SRN17812RWY 35RminA1A4A10- maxαAβA*Time windowfor the travel time in each air segment[min travel time; max travel time]
15AG Model Exit due date γA (γA = - planned landing time) A1 A4 A10 A13 CommonGlide PathHolding CirclesAirSegmentsRunwaysAA1TOR410131659RWY 35LMBR2611157143SRN17812RWY 35RA1A4A10A13A15A16AOUTαAγAβA*Exit due date γA(γA = - planned landing time)
16AG Model Potential conflict on resource 15 ! Holding CirclesAirSegmentsCommonGlide PathRunwaysAA1TOR410131659Potential conflicton resource 15 !RWY 35LMBR26111571417B3SRN812BRWY 35RA1A4A15A10A13AOUTA16*B3B8B15B12B14BOUTB17αAαBβAγAβBγBAircraft ordering problem between A and B on the common glide path (resource 15) : Longitudinal and diagonal distances have to be respected
17AG ModelHolding CirclesAirSegmentsCommonGlide PathRunwaysAA1TORAircraft ordering problem between B and C for the runway (resource 17): This is a no-store resource!410131659RWY 35LMBR261115C71417B3SRNC812BRWY 35RA1A4A15A10A13AOUTA16*B3B8B15B12B14BOUTB17αAαBβAγAβBγBCOUTC17γCαC
18Modeling a Terminal Control Area Solution Framework and Algorithms Presentation outlineIntroductionModeling a Terminal Control AreaSolution Framework and AlgorithmsComputational ExperimentsConclusions and Ongoing ResearchThis work was partially supported by the Italian Ministry of Research, project FIRB “Advanced tracking system in intermodal freight transportation”.
19Developing a decision support tool From a logical point of view, ATC decisions can be divided into:• Routing decisions, where a route for each aircraft has to be chosen in order to balance the use of TCA resources.• Scheduling decisions, where routes are considered fixed,and feasible aircraft scheduling solutions have to be determined.In practice, the two decisions are taken simultaneously.However, the main objective of real-time routing decisions is typically to balance the use of alternative runways and airways while that of real-time scheduling is the delay minimization.19
20MILP (Mixed-Integer Linear Programming) model FIXED AIRCRAFT ROUTES20
21MILP (Mixed-Integer Linear Programming) model FLEXIBLE AIRCRAFT ROUTES ns: number of routes of aircraft sna: number of aircraft21
22Rolling Horizon (RH) approach Time horizon T1Roll periodTime horizon T2Roll periodTime horizon T3timeLength of the overall traffic prediction horizon
24RH: Stage 2 A1 A4 A10 A13 A15 A16 AOUT B14 B15 B17 BOUT C17 COUT AirSegmentsHolding CirclesCommonGlide PathRunwaysAA1TOR410131659RWY 35LMBR261115CRoll Period = 5Time horizon T2 [5;20]714C17B3SRN812BRWY 35RA1A4A10A13A15A16AOUTαA = 10βA = -10αB = 5*B14B15B17BOUTNon so se mettere la observation. Sono indeciso perché:Da un lato spiega come rendiamo il modello dinamicoDall’altro non è poi messa negli esperimenti (per motivi di tempo mi pare abbiamo deciso di escluderlo)αC = 17βB = -5C17COUTObservation: At this stage the release time of A and C can be updated dynamically if updated entry times are knownβC = -17
25Decision Support System based on the Rolling Horizon Approach InstanceGeneratorFeasible SolutionSet new roll periodAircraft notfully processedSingle StageSolverAircraft entry times (dynamic information)XML fileAirport ResourcesAircraft TimesAircraft RoutesTime HorizonRoll period(if any)L’approccio mi permette di: 1) scomporre il problema in sottoproblemi (multi stage approach)2) tener conto di nuove informazioni/misure del tempo stimato di ingresso nella TCA
27Scheduling heuristics FCFS (First Come First Served)Gives precedence to the first aircraft requiring the resource.Greedy heuristics (Pranzo et al. 2003):AMCC (Avoid Most Critical Completion time)Chooses the pair containing the alternative arc which wouldcause the largest increase in consecutive delay.AMSP (Avoid Most Similar Pair)Chooses the pair with the largest sum of consecutive delays.JGH (Job Greedy Heuristic) Selects all the alternative arcs involving a chosen aircraft, so that the sequencing of all the operations of this aircraft are fixed in one step.
28Branch and bound (BB) algorithm Lower bound: Blocking extension of the Jackson pre-emptiveschedule [Carlier & Pinson 1989] on air segments and runways.Constraint propagation: Speed up based on network topology and graph proprieties developed in [D’Ariano et al. 2007].Experimental setup:Branching rule: Choose the most critical unselected alternativepair with criteria AMSP and branch on this pair.Hybrid search strategy: Alternate four repetitions of the depth-first visit with the choice of the open node of the search tree withthe smallest lower bound among the last five generated nodes.Settaggio sperimentale + proprietà delle implicazioni statiche e dinamiche
29ASP with flexible routes: Move & neighbour We start from the solution obtained for the ASP problem withfixed routes. The search for better aircraft routes is as follows:A move is to change one aircraft route and its evaluation is to solve the associated ASP problem with fixed routes;At each iteration the best (local) move is taken from a set of neighbours of a current ASP solution;Neighbourhood: It is well known that a job shop scheduling solution can be improved by changing the critical path C(S) related to the current graph selection (solution) S only [Balas OR 69];We use ramified critical paths [D’Ariano et al. 2008].
31ASP with flexible routes: Tabu Search (TS) The ramified critical paths are well focused on reducing the maximum consecutive delay but, in general, are not opt-connected[Corman et al. 2010].Example:Critical path on aircraft B onlybut rerouting aircraft A or Cmay reduce the critical pathA tabu search algorithm is proposed to escape from local minima by taking a non-improving move and then forbidding the inverse move for a given number of iterations.Another technique to escape from local minima is to perform moves based on restart technique.
32Processor Intel i7 (2.84 GHz), 8 GB Ram Presentation outlineIntroductionModeling a Terminal Maneuvering AreaSolution Framework and AlgorithmsComputational ExperimentsConclusions and Ongoing ResearchProcessor Intel i7 (2.84 GHz), 8 GB RamThis work was partially supported by the Italian Ministry of Research, project FIRB “Advanced tracking system in intermodal freight transportation”.
33Centralized vs Rolling Horizon 3-hour horizon[20 instances]
34Static/Dynamic Case: BB vs FCFS 1-hour horizon[20 instances]
35Modeling a Terminal Control Area Solution Framework and Algorithms Presentation outlineIntroductionModeling a Terminal Control AreaSolution Framework and AlgorithmsComputational ExperimentsConclusions and Ongoing ResearchThis work was partially supported by the Italian Ministry of Research, project FIRB “Advanced tracking system in intermodal freight transportation”.
36Achievements firstname.lastname@example.org Detailed ASP models have been investigated for MXP and FCO;The computational experiments proved the effectiveness of our rolling horizon approach compared to a centralized approach;Optimization algorithms outperforms simple rules, both for static and dynamic cases, in terms of delay and travel time minimization;The BB-based rolling horizon approachsolves the one-hour instances quickly.
37Further research directions Evaluation of aircraft rescheduling and rerouting approaches for optimal decision making in case of various traffic disturbancesStudy of multiple criteria for aircraft traffic control at busy TCAs(e.g. delay, priority, fairness, environmental and other cost factors)Development of detailed models for the coordination & real-time optimization of en-route, approach and TCA traffic managementTransformative: Practical realization of integrated (closed-loop) intelligent decision support systems at traffic control centers
38Malpensa Airport: Results from D’Ariano et al. 2012 AlgorithmTimeHorizonMax ConsDelay (s)Avg ConsMax TotAvg TotDelayedAircraftTotal DTTS (s)BB30139.124.1659120.15.81832TS (C)91.612.6622107.36.51587TS (A)127.822.1655115.27.41868TS (R)94.914.1635108.86.8167245305.961.61227253.413.2364616635.91087200.613.52988234.957.91125229.814.9170.731.61104200.02901RESULTS ON THE AIRCRAFT REROUTING MODULE: The Tabu Search (TS) algorithm is applied to perform three kinds of optimal rerouting strategies: Combined (C), Air (A) and Runway (R).BEST DSS CONFIG: Compared to the BB for aircraft scheduling, TS (C) achieves an improvement on the max cons delay minimization by 34% for 30-min instances and by 46% for 45-min instances.