CDM in Departure Sequencing with Adapted Rubinstein Protocol Vitor - - PowerPoint PPT Presentation
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM in Departure Sequencing with Adapted Rubinstein Protocol Vitor Filincowsky Ribeiro University of Bras lia UnB Computer Science Department
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks
Vitor Filincowsky Ribeiro
University of Bras´ ılia – UnB Computer Science Department – CIC Air Transportation Research Lab – Translab
Bras´ ılia, August 21th 2015
Vitor F. Ribeiro CoDMAN 21/08/2015 1 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks
1 Introduction 2 Theoretical Foundations 3 CoDMAN 4 Simulations and Results 5 Conclusion remarks
Vitor F. Ribeiro CoDMAN 21/08/2015 2 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Vitor F. Ribeiro CoDMAN 21/08/2015 3 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Significant increase of pax movement in the Brazilian airports
Challenge: prevent congestion in the terminals Bottleneck: application of constraining measures Consequences: increase of costs associated to delays
Vitor F. Ribeiro CoDMAN 21/08/2015 4 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
TWR Operations Departing order and separation time between aircraft Auxiliary computer systems No decision support so far Results depend exclusively on the controller’s expertise
Vitor F. Ribeiro CoDMAN 21/08/2015 5 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Necessity of an efficient departure management Intelligent slot allocation Dynamic adjustments in the departure queue Fairness in delay costs distribution Bargaining Complex procedure
Vitor F. Ribeiro CoDMAN 21/08/2015 6 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Negotiation among the agents: no documented slot allocation solution
Vitor F. Ribeiro CoDMAN 21/08/2015 7 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Negotiation among the agents: no documented slot allocation solution Proposal: computational model for departure management in the airports Game Theory Collaborative Decision Making (CDM) Collaborative Departure Management - CoDMAN
Vitor F. Ribeiro CoDMAN 21/08/2015 7 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Major objective Provide a computational model for an efficient departure management in the airports
Vitor F. Ribeiro CoDMAN 21/08/2015 8 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Major objective Provide a computational model for an efficient departure management in the airports What do we need: Development of a system which implements CDM for aircraft queuing prior departure
Vitor F. Ribeiro CoDMAN 21/08/2015 8 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Simulation of critical scenarios found in the airports under distinct demand conditions
Vitor F. Ribeiro CoDMAN 21/08/2015 9 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Simulation of critical scenarios found in the airports under distinct demand conditions Development of a CDM-based queuing algorithm
Vitor F. Ribeiro CoDMAN 21/08/2015 9 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Motivation Objectives
Simulation of critical scenarios found in the airports under distinct demand conditions Development of a CDM-based queuing algorithm Development of a slot allocation methodology using negotiation among the agents in Game Theory
Vitor F. Ribeiro CoDMAN 21/08/2015 9 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Vitor F. Ribeiro CoDMAN 21/08/2015 10 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Collaborative Decision Making Joint operation among all impacted entities Fairness in cost distribution Sharing of complete and up-to-date information
Vitor F. Ribeiro CoDMAN 21/08/2015 11 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Airport Collaborative Decision Making Capacity and traffic flow management in the airport scope Delay reduction and increase of predictability of events Resource usage optimization
Vitor F. Ribeiro CoDMAN 21/08/2015 12 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Mathematical approach for interest conflicts or cooperation in interactive situations Individuals or sets (agents) in a game are entities that should adopt a given behaviour Status progression consists in agents decisions or in results of probabilistic events
Vitor F. Ribeiro CoDMAN 21/08/2015 13 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Formal description of strategic situations Actions of the agents are interdependent Randomness of agents decisions is granted by mixed strategies
Vitor F. Ribeiro CoDMAN 21/08/2015 14 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Cooperative vs Non-cooperative games
Vitor F. Ribeiro CoDMAN 21/08/2015 15 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Cooperative vs Non-cooperative games Every situation can be mapped into a real number (payoff ) which express the players’ preference or interest Payoff function
Vitor F. Ribeiro CoDMAN 21/08/2015 15 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Cooperative vs Non-cooperative games Every situation can be mapped into a real number (payoff ) which express the players’ preference or interest Payoff function The player’s goal is maximize its payoff function The next scenario should be as favourable as possible
Vitor F. Ribeiro CoDMAN 21/08/2015 15 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
All players act in order to maximize their payoff Agents observe or try to predict the actions of the other players Every choice made by a player is the best response given the choices of the other players A player cannot unilaterally alter its strategy and obtain a greater payoff Central solution in strategic equilibrium
Vitor F. Ribeiro CoDMAN 21/08/2015 16 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Negotiation among agents Complete information about the actors Alternate offers protocol Iterations consist in the evaluation of the opponent’s offer Acceptance, new offer proposition, negotiation abandonment
Vitor F. Ribeiro CoDMAN 21/08/2015 17 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CDM Game Theory
Negotiation among agents Complete information about the actors Alternate offers protocol Iterations consist in the evaluation of the opponent’s offer Acceptance, new offer proposition, negotiation abandonment Final offer arbitration Unsuccessful negotiation Control entity applies one of the previous valid offers
Vitor F. Ribeiro CoDMAN 21/08/2015 17 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Vitor F. Ribeiro CoDMAN 21/08/2015 18 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Main tasks Collaborative functionality under CDM principles Framework for real-time attendance and tactical decision support about usage of resources on ground Search for an optimum departure sequence by the aircraft
Vitor F. Ribeiro CoDMAN 21/08/2015 19 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Access to operational attributes of the aircraft Interface with real-time data provision systems Exploitation of agents’ rationality Part of the controller’s responsibility is transferred
Vitor F. Ribeiro CoDMAN 21/08/2015 20 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Departure sequencing is a basic allocation problem Each agent must be provided complete and up-to-date information Agents are the aircraft which negotiate with each other the slot occupancy In ATFM context, flights are agents that work in order to maximize their efficiency
Delay costs reduction Individual actions conditioned to behaviour of other agents
Vitor F. Ribeiro CoDMAN 21/08/2015 21 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Initial departure queue is established according to the flight plans Manipulations in the queue take advantage of free slots and try to preserve the primary departure order Responsive actions
Vitor F. Ribeiro CoDMAN 21/08/2015 22 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Environment composed by n players Specific payoff function
Vitor F. Ribeiro CoDMAN 21/08/2015 23 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Environment composed by n players Specific payoff function Nash Equilibrium convergence Global stability point Isolated actions do not yield better individual results
Vitor F. Ribeiro CoDMAN 21/08/2015 23 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Two main categories of agents Aircraft: search for the accomplishment of the scheduled take-off and landing times Airport: prevention of saturation situations conditioned to current capacity and safe operation of all flights
Vitor F. Ribeiro CoDMAN 21/08/2015 24 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Operational parameters of airport and TMA
F: set of active flight in airport and TMA aircraft on ground (Fg), landed included (F −
g )
airborne aircraft (Fa), departed included (F −
a )
Ip: airport capacity parking slots (I f
p )
runways (I r
p)
Ca(t): landing capacity in instant t Cd(t): departing capacity in instant t Qa: set of aircraft in TMA waiting authorization for landing Qd: set of aircraft in land waiting authorization for take-off K: total delay cost
Vitor F. Ribeiro CoDMAN 21/08/2015 25 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Departing capacity is defined by the amount of aircraft that can be delivered Cd(t) = I r
p −
Ia(t) + Id(t)
Vitor F. Ribeiro CoDMAN 21/08/2015 26 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Departing capacity is defined by the amount of aircraft that can be delivered Cd(t) = I r
p −
Ia(t) + Id(t)
Landing capacity is defined by airport instant capacity Ca(t) = min{I f
p − F − g , I r p −
Ia(t) + Id(t) }
(2)
Vitor F. Ribeiro CoDMAN 21/08/2015 26 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Departing capacity is defined by the amount of aircraft that can be delivered Cd(t) = I r
p −
Ia(t) + Id(t)
Landing capacity is defined by airport instant capacity Ca(t) = min{I f
p − F − g , I r p −
Ia(t) + Id(t) }
(2) Total delay cost is defined by the sum of the individual costs K =
n
ki (3)
Vitor F. Ribeiro CoDMAN 21/08/2015 26 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Aircraft attributes
td
i : original departure time allocated to an aircraft;
ta
i : calculated arrival time at the destination airport;
vi: cruise speed; tf
i : estimated flight time until arriving to the final destination;
ci: aircraft max occupancy, in number of passengers; Di: maximum delay acceptable to an aircraft, expressed in minutes; pi: number of passengers onboard an aircraft; Bi: size of an aircraft, expressed as a real number in [1.0, 3.0]; wi: aircraft wingspan; σi: significance of an aircraft, expressed as a positive real number; li: the actual delay imposed on an aircraft, expressed in minutes; ki: actual weighted individual cost, i.e., the payoff function of the aircraft.
Vitor F. Ribeiro CoDMAN 21/08/2015 27 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Aircraft significance is defined by pax, size and distance to the destination airport σi = vitf
i
Bi wi pi ci (4)
Vitor F. Ribeiro CoDMAN 21/08/2015 28 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Aircraft significance is defined by pax, size and distance to the destination airport σi = vitf
i
Bi wi pi ci (4) Weighted cost of individual delay is based on the aircraft significance ki = li Di σi (5)
Aircraft with greater significance are those which present greater operational costs
Vitor F. Ribeiro CoDMAN 21/08/2015 28 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
First moment: FIFO order Scenario assessment Opportunity to adjust the departure slots according to individual interests Knowledge about the other agents favours the cooperative behaviour
Vitor F. Ribeiro CoDMAN 21/08/2015 29 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Basic slot allocation flowchart
Vitor F. Ribeiro CoDMAN 21/08/2015 30 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Possible actions during the negotiation Advance (adv): an earlier slot is offered to the opponent Delay (del): opponent receives the offer of delaying its push-back time in order to relinquish its slot to the proponent Swap (swp): opponent is asked to swap its slot with the
this decision
Vitor F. Ribeiro CoDMAN 21/08/2015 31 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Possible answers to the offers Accept (acc): opponent accepts the proposal and earns the proper payoff Reject (rej): opponents rejects the offer and both shall accept the arbiter intervention Counter-offer (ctr): opponent calculates another counter-offer, restarting the bargaining process and turning into the active agent
Vitor F. Ribeiro CoDMAN 21/08/2015 32 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Departure queue checkup process for negotiation
Vitor F. Ribeiro CoDMAN 21/08/2015 33 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Eagerness gradient γ = w max(w, tobt − tcur) (6)
Vitor F. Ribeiro CoDMAN 21/08/2015 34 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Eagerness gradient γ = w max(w, tobt − tcur) (6) Discount factor δ = 1 + γ1γ2 (7)
Vitor F. Ribeiro CoDMAN 21/08/2015 34 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks CoDMAN Operation Game Theory in CoDMAN System Implementation
Rubinstein model adapted to CoDMAN
Vitor F. Ribeiro CoDMAN 21/08/2015 35 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Vitor F. Ribeiro CoDMAN 21/08/2015 36 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Bras´ ılia International Airport (SBBR) Yard: 40 aircraft Runways for departure and landing: 2 Both runways can be simultaneously used
Vitor F. Ribeiro CoDMAN 21/08/2015 37 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
FIRs in Brazil (FIR-BS evidenced)
Vitor F. Ribeiro CoDMAN 21/08/2015 38 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Attribute Before After Flights out of schedule 35 36 Avg delay for delayed aircraft 3.48 minutes 3.26 minutes Avg cost for delayed aircraft 643.95 476.193 Total delay 73 minutes 62 minutes Std deviation for delays 0.92 1.05 Std deviation for delay costs 229.17 182.43 Total delay costs 13522.9 9047.67 Avg cost per delayed aircraft decreased 26% Total scenario cost decreased 33%
Vitor F. Ribeiro CoDMAN 21/08/2015 39 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Attribute Before After Ground Holding measures 7 6 Airborne Holding measures 2 2 Flights out of schedule 43 45 Avg delay for delayed aircraft 2 minutes 1.66 minutes Avg cost for delayed aircraft 385.884 272.286 Total delay 86 minutes 73 minutes Std deviation for delays 0.98 1.04 Std deviation for delay costs 264.16 199.41 Total delay costs 16593 11980.6 Total scenario cost decreased 27.8%
Vitor F. Ribeiro CoDMAN 21/08/2015 40 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Virtual reduction of airport infrastructure to only one runway Impact Qtd. Departures 201 Landings 229 Total flights processed 515 Ground Holding application 68 Airborne Holding application 32 Delayed flights 102 Airport infrastructure quickly reaches its maximum operation capacity Scenario degradation is quite evident
Vitor F. Ribeiro CoDMAN 21/08/2015 41 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Evaluation of degradation when one runway is removed Attribute Before After Avg delay for delayed aircraft 3.23 minutes 3.01 minutes Avg cost for delayed aircraft 595.233 544.848 Total delay 329 minutes 319 minutes Std deviation for delays 2.34 2.39 Std deviation for delay costs 570.475 523.71 Total delay costs 60713.8 56119.4 Avg increase of 65% in individual delay costs Scenario performance is almost 4x worse in terms of operational costs
Vitor F. Ribeiro CoDMAN 21/08/2015 42 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Aerodrome operates with three virtual runways Impact Qtd. Departures 202 Landings 231 Total flights processed 518 Ground Holding application Airborne Holding application Off-schedule flights 35 There was no need of constraining measures application at any time
Vitor F. Ribeiro CoDMAN 21/08/2015 43 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Scenario description Simulation scenarios
Scenario optimization when a runway is added Attribute Before After Avg delay for delayed aircraft 2.09 minutes 1.69 minutes Avg cost for delayed aircraft 396.14 277.27 Total delay 73 minutes 61 minutes Std deviation for delays 0.93 0.99 Std deviation for delay costs 243.83 192.33 Total delay costs 13865 9981.53 Avg scenario cost decreased 28%
Vitor F. Ribeiro CoDMAN 21/08/2015 44 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
Vitor F. Ribeiro CoDMAN 21/08/2015 45 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
Simulation Cost Cost Optimization scenario Before After (avg / total) Static allocation 13522.9 9047.67 26% / 33% Dynamic allocation 16593 11980.6 29.4% / 27.8% Runway reduction 60713.8 56119.4 8.5% / 7.6% Runway enhancement 13865 9981.53 30% / 28%
Vitor F. Ribeiro CoDMAN 21/08/2015 46 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
Static sequencing
Decrease of 33% in sequencing costs evinces the benefits of an adequate strategic planning
Simple dynamic allocation
Optimization: 27.8% Standard deviation of the costs presented the greater reduction, evincing that this allocation is the one which distributes the delay costs among the aircraft in the better way
Vitor F. Ribeiro CoDMAN 21/08/2015 47 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
Capacity reduction
Loss of queue maneuverability Free slots go scarce as time goes by Increase of swap offers Optimization was only 7.6% Bad cost distribution
Capacity augmentation
Optimization: 28% (best result) Despite it is a good solution, it is not realistic
Vitor F. Ribeiro CoDMAN 21/08/2015 48 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
Objectives The proposed objectives were accomplished Conclusion: a scenario modeled under the principles of CDM in fact behaves as a complete information cooperative game
Vitor F. Ribeiro CoDMAN 21/08/2015 49 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
1 Innovative solution to slot allocation problem Vitor F. Ribeiro CoDMAN 21/08/2015 50 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
1 Innovative solution to slot allocation problem 2 Aid to the decision making process by the human agent in a
complex environment
Vitor F. Ribeiro CoDMAN 21/08/2015 50 / 51
Introduction Theoretical Foundations CoDMAN Simulations and Results Conclusion remarks Evaluations Research evaluation
Vitor F. Ribeiro CoDMAN 21/08/2015 51 / 51