Air Transportation Network Load Balancing using Auction-Based Slot

Allocation for Congestion Management

GL Donohue, L Le, CH Chen, D WangGL Donohue, L Le, CH Chen, D WangCenter for Air Transportation Systems Research

Dept. of Systems Engineering & Operations ResearchGeorge Mason University

Fairfax, VANEXTOR NEXTOR WyeWye River ConfRiver Conf

June 21June 21--23, 200423, 2004

Outline

Description of the Congestion Problem− Chicago O’Hare Airport− NY La Guardia Airport

History of Congestion Management in the USAuction model for airport arrival slotsChicago ORD airport case study− simulated scenarios− results and interpretation

Observations and Recommendations

National Airspace System Characteristics

The NAS is a Stochastic AdaptiveStochastic Adaptive Network−− Stochastic:Stochastic: The system is characterized by PDF’s−− Adaptive:Adaptive: These PDF’s are a function of the System State and

Airline Market Decisions

Reasons that the NAS Cannot be DeterministicCannot be Deterministic:− Weather (winds, hazardous weather)− Mechanical Equipment Characteristics− Air Traffic Control System (including Controllers)− Aircraft Control System (including Pilots)− Airline Schedules set by varying Market Conditions

All Analysis and FAA Rules Must Acknowledge this Fundamental Nature in the Future

Operations are Back but ORD Delays are Worse

Monthly Total Operations at Major US Airports (Source: ETMS)

60,000

65,000

70,000

75,000

80,000

85,000

90,000

Jan-0

1Mar-

01May

-01Ju

l-01

Sep-01

Nov-01

Jan-0

2Mar-

02May

-02Ju

l-02

Sep-02

Nov-02

Jan-0

3Mar-

03May

-03Ju

l-03

Sep-03

Nov-03

Jan-0

4

Month

#Flights

ATL ORDAir travel is gradually picking up

Delayed Flights of Major US Air Carriers at ORD (Source: BTS)

02,0004,0006,0008,000

10,00012,00014,000

Jan-01

Mar-01

May-01

Jul-0

1Sep-01Nov-0

1Ja

n-02Mar-0

2May-0

2Ju

l-02

Sep-02Nov-0

2Ja

n-03Mar-0

3May-0

3Ju

l-03

Sep-03Nov-0

3Ja

n-04

Month

#Flights

Departure Arrival

Congestion is coming back

Chicago O’Hare: A Maximum Capacity Hub Airport

Runway layout:Departure/Arrival Pareto Trade-off:FAA/DoT 2001 Benchmark Report ASPM Data April 2000 VMC

IMC Capacity

ORD in Dec 2003:Result of HDR Removal

DoT/FAA 2001 Benchmark Report:Data April 2000 VMC

ASPM Data Dec 2003

Scheduled Arrival Rate for Different Airlines at ORD in Nov. and Dec. 2003

Scheduled Arrival Rate (15 min time block)

Time (15 min time block)

# of

sch

edul

ed a

rriv

als

• The Green line = 21 arr. /15 min, upper-bound of IMC AR at ORD in Nov. and Dec. 2003

• The Red line = 25 arr./15 min, Max AR from FAA Benchmark Report for all arrivals

Data from BTS which only includes domestic flight data for 15 certificated airlines

Smaller Aircraft trend Exacerbates Congestion

Frequency competition Reduces Seat Capacity and Increases FAA Operational Load

O’Hare Airport Yearly Throughput

911,917

922,817928,691

1998 1999 2000 2001 2002 2003Year

Flight Operations

896,104 896,228

908,989

72,501,988 72,610,12172,145,489

67,448,06466,565,952

69,508,672

Passengers

Small aircraft make inefficient use of runway capacity: 50% Flights Provide 70% Enplanement Opportunities

ORD Scheduled Operations (BTS Dec 2003)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

cumulative flight share

seats/aircraft

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1cumulative seat share

0

50

100

150

200

250

300

350

400

450

500

Enplanement Capacity vs. Operational Capacity

NY LaGuardia: A Maximum Capacity non-Hub Airport

1 Arrival Runway1 Departure Runway45 Arrivals/Hr (Max)80 Seconds Between Arrivals11.3 minute Average Delay77 Delays/1000 Operations40 min./Delay

TOTA L SC HED U LED OPER A TION S A N D C U R R EN T OPT IM U M R A TE B OU N D A R IES

0

4

8

12

16

20

24

28

32

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Schedule Facility Est. M odel Est.

NY LaGuardia :A Maximum Capacity Non-Hub Airport

0

10

20

30

40

50

60

0 10 20 30 40 50 60

Departures per Hour

Arriv

als

per H

our

ASPM - Apr 2000 - Visual Approaches

ASPM - Oct 2000 - Visual Approaches

Calculated VMC Capacity

Optimum Rate (LGA)

40,40

Each dot represents one hour of actual traffic

during April or October 2000

Runway layout:

Departure/Arrival Pareto Trade-off:

ASPM Data April 2000 VMC

IMC Capacity

Factors that Determine Capacity

Local Airport Authority− #Runways, #Taxiways, High-speed turnoffs, #Gates,

RW spacing, RW configuration, Noise Restrictions, etc.FAA− ATM/CNS Equipage, Separation Standards, ATC

Procedures and Airspace Design Weather− Winds, Ceiling, Visibility, Severe weather

Airline Schedules− Network Banking Requirements− Market Competition Strategies

History of US Slot Management

-Limited #IFR slots during specific time periods-Negotiation-based allocation-ORD, LGA, DCA,and JFK

1968

High-Density Rule

Apr2000

Exempted from HDR certain

flights to address

competition and small market

access

AIR-21

Jan2001

Lottery

1978

Deregulation

Use-it-or-lose-it rule based on 80% usage

1985

Slot ownership

2007

End of HDR. What’s next?

-Congestion pricing?-Auction?

LGA Airport Slot Control

Jun2002

ORD Airport Slot Control

HDRremoved

Jan2004

FAA-regulated5% reduction

June2004

FAA-regulated2.5% reduction

DOT’s congestion management

2007

LGA Airport Slot Control

Jun2002

ORD Airport Slot Control

HDRremoved

Apr2000

Exempted from HDR certain

flights to address

competition and small

market access

AIR-21

Jan2001

“Slottery”

Nov2000

Major chokepoint300 new daily flights25% total network delay

Jan2004

FAA-ordered 5% reduction by AA and UA

Sep 2001

Oct 2001

Nov2003

130 more flights42% delayed (from 18% in 2002)13 mishaps (from 3 in 2002)

June2004

FAA-regulated 2.5% reduction by AA and UA

Over-Scheduling

Over-Scheduling

End of HDR

Congestion Management Approaches

Administrative− negotiation-based IATA biannual conferencesMarket Based− weight-based landing fee: no incentive for large

aircraft – inefficient Enplanement capacity− time-based congestion pricing: not reveal the true

value of scarce resources− DoT/FAA supervised MarketMarket--based Auctions based Auctions of

Arrival Metering-Fix Time SlotsHybrid

Auction Model Design Issues

Feasibility− Package slot allocation for arrival slots− Politically acceptable net prices

Optimality− Efficiency: i.e. Match Customer value to Cost

Maximum Schedule PredictabilityOptimum airline schedule and aircraft assignmentMinimum passenger ticket price

− Regulatory standards: capacity, international flight priorities− Equity:

Stability in scheduleAirlines’ need to leverage Prior InvestmentsAirlines’ competitiveness : new-entrants vs. incumbents

Flexibility− Primary market at strategic level − Secondary market at tactical level

Design Approach

Objective:− Obtain Better Utilization of Nation’s Airport Network

Infrastructure – Network Load BalancingNetwork Load Balancing− Provide Cities an Optimum FleetOptimum Fleet Mix − Ensure Fair Market AccessFair Market Access Opportunity− Increase Schedule PredictabilitySchedule Predictability - reduced queuing delays

Assumptions− Airlines will make optimum use of slots they license

Auction rules: Bidders Could Be Ranked using a linear combination of: − Monetary offer (combination of A/C equipage credit and cash)− Flight OD pair (e.g. international agreements, etc.)− Airline’s prior investment ?− On-time performance ?

Strategic Auction Analytical Approach

NAS

Auction Model

Schedules

Analysis & Feedback

Bids

SlotsAirlines Auctioneers

Network Model

-Auctions only at Capacitated Airports

-Auction Licenses good for 5 to 10 years

Network Model used to Evaluate Auction Effectiveness

LGAORDMSP

DTW

ATLDFW

LAXIAD

BWIDEN

PHX

SFO

13-node network

departure separation

arrival separation

Trailing aircraft Small Large B757 HeavyLeading aircraftSmall 2.5/80 2.5/68 2.5/66 2.5/64Large 4/164 2.5/73 2.5/66 2.5/64B757 5/201 4/115 4/102 4/101Heavy 6/239 5/148 5/136 4/104

Runway capacity determined by♦ Wake Vortex Separation Standards (nmiles/seconds) (M. Hanson)

♦ and a scale factor to account for runway dependency

Auction Model Process

More demand than capacity

Raise price clock by a fixed increment

End auction process

Submit slot demands and flight

info.

Yes

No

Auctioneer’s actionAirline’s action

Call for demands

An adaptation of Clock Combinatorial Auction Model (Porter, D., Rassenti, S., Roopnarine, A., and Smith, V.)

Set initial price for each 15-min

interval

Clock price raised?

For each auctioned interval

NoYes

Solve combinatorial

package bidding

ORD Simulated Arrivals in VMC – No Auction

Domestic Arrivals (Dec 2003)

0

10

20

30

40

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

#Flights Scheduled (BTS) simulated mean

0

5

10

15

20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Time

Min

Mean of Estimated Average Arrival

Delay (good weather)

Scheduled Domestic Arrivals at ORD

010203040

0 3 6 9 12 15 18 21

Time

Min

Before auction After auction auctioned at 18ops

ORD Simulated Schedule and Delay: 3 Auctioned Capacity Limits (VMC-IMC)

Estimated Runway Queuing Delay

0

5

10

15

0 3 6 9 12 15 18 21

Time

Min

Before auction After auction auctioned at 18ops

21 Ar/15 min

18 Ar/15 min

De-peaking results in Significantly Improved Passenger Schedule

Predictability

0

20

40

60

80

100

120

140

160

0 20 40 60 80 100 120 140 160

Departures per Hour

Arr

ival

s pe

r Hou

r

ASPM - April 2000 - Instrument Approaches

Calculated IMC Capacity

ASPM - April 2000 - Visual Approaches

Optimum Rate (ORD)

100,100

80,80

Calculated VMC Capacity

Average Estimated Flight Delay

0

50

100

150

200

250

300

350

400

14 15 16 17 18 19 20 21

Quarterly Airport Arrival Rate in Different Weather Conditions

Min

Current Schedule

auctioned at 21 ops/15 min

auctioned at 18 ops/15 min

IMCVMC

Auction Produced Rolling Banks Changes the Distribution of Delays

3 8 13 180.00

0.05

0.10

0.15

0.20

0.25

Estimated Arrival Queuing Delay

min

Current ScheduleAuction-produce schedule

Probability of Delay Duration

ORD Flights to be Rerouted vs. Average Actual Cancellations / day

Estimated #flights to be rerouted

050

100150200250

300350

14 15 16 17 18 19 20 21

Quarterly Airport Arrival Capacity in Different Weather Conditions

#Flights

18 ops/15 min

21 ops/min

Current Schedule

Daily number of cancellations (BTS)

0

50

100

150

200

250

12.2003 1.2004 2.2004 3.2004

26 29 2926

8773

Significantly Improved ORD Passenger Schedule Predictability

Auction Produced a Coordinated Airline De-Peaked ScheduleSimulated ORD Auction at 21 Arrivals/15 Min:− Reduced Delays by Over 80%− Required only 26 Flights to be Re-Scheduled through

another Non-Capacitated Hub Airport− This Reduction is Comparable to the Reported Daily

Flight Cancellation Rate

Research Issues to be Addressed

Who is Eligible to Bid?− Airlines, Airports, General Aviation, Investors

What is the Fundamental Bidding Metric?− $/15 min Slot @ 95% Confidence, $-Passenger/Aircraft Slot…

How Many Slots Should be Auctioned (arr @Prob. Delay (min))?− VMC ROT @ N(4,22), IMC WV @ N(8,42), IMC WV @ N(15,82) …

What Bid Combinations will be Allowed?− Packages w/ Ranked Priorities, Intercity Packages, etc.− Bidding Activity Rules

What are the Payment Options?− Up Front for X yr. Lic., Monthly Royalty Payments for X Yrs.

Who gets the Money?− Airports (PFC Sub.), Airlines (Equip. Vouchers), FAA (Ticket Tax

Sub.)What are the Secondary Market Rights?What is the Winner Determination Algorithm?Auction Frequency/Duration of Slot License?− License for 5 yr., 10 yr., ?

Observations on Research to Date

Combinatorial Clock Auctions Offer a Promising Market-Based approach to Congestion ManagementAuction Proceeds could be used as Incentives to the Airports for Infrastructure Investments and to the Airlines for Avionics InvestmentsCongestion Management at Critical Network Node Airports will have a Profound Effect on Increasing Passenger Travel PredictabilitySimple Auctions might Exclude Small Aircraft and/or Small markets from Hub Airports− Simple Bidding Rules can Prevent this Problem

Future work

Conduct 3 FAA Strategic Simulations to Resolve Slot Allocation Issues − First Simulation would Examine a Variety of Policy

Problems/Options (Include a broad collection of Stakeholders)

− Second Simulation would examine specific sets of auction rules and instruments

− Third Simulation would use Results of first two to Evaluate Modified Congestion Mgt. Options

Continue Model development to Refine Combinatorial Package Bidding Simulations to Evaluate Proposed Auction Rule Set

Backup

Simulation Model for Testing Auction Design

General Assumptions:− Aircraft can arrive within allocated 15 min. Arrival Time slots with

Required Time-of-Arrival errors of 20 seconds (using Aircraft RTA Capabilities)

− Auction items: Metering Fix Arrival Slots in 76 15-min bins (5:00am till 24:00am) up to 21 arrivals/bin

Input:− Dec 2003 BTS schedule of 2186 flights domestic flights to ORD (80% of

total traffic)ORD Scheduled Arrivals (Source: ASPM, BTS Dec 2003)

0

10

20

30

40

50

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Time

#Ops

Total ArrivalsDomestic Arrivals of 15 certificated airlines

Airline model assumptions

Single market, single item bidding

Airlines’ flexibility for changing schedule: one 15-min bin

Homogenous and honest bidding with upper threshold proportional to aircraft size

original scheduled 15-min interval

15min15min

bids withdrawn

bids withdrawn

u(bi) = 1

0bi bi+1

Linear decreasing

bi-1

Airline Package Bidding Model

bilb ≤ bi ≤ bi

ub

Δbi,i+1lb ≤ bi+1 – bi ≤ Δbi,i+1

ub

Possible packages Pjk for Pj:

Target Slots Pj: b1 b2 u(Pj1) = 1

PjkPj

1 Pj2

Utility Function

u(Pjk) = 0

LP Model:Maximise

Variables:

∑ ∀≤k

kj jx 1

∑∑ ⋅j k

kj

kj xPu )(

Subject to:

BxPj k

kj

kj ≤⋅⋅Π∑∑

highly flexibleschedule

non flexibleschedule

⎩⎨⎧

=01k

jxif airline bids for package Pj

k

{Pjk} set of package bids

B airline bidding budget

Π price vectorotherwise

Network Simulation Model used to Evaluate Auction Effectiveness

Trailing aircraft Small Large B757 HeavyLeading aircraftSmall 2.5/80 2.5/68 2.5/66 2.5/64Large 4/164 2.5/73 2.5/66 2.5/64B757 5/201 4/115 4/102 4/101Heavy 6/239 5/148 5/136 4/104

– Stochastic Queuing Model– 12 Capacitated Airports–1 Airport Unconstrained sink and source– ORD Runway capacity determined by

Wake Vortex Separation Standards (nmiles/seconds) (M. Hanson)

and a scale factor to account for runway dependencyweather effect

0

20

40

60

80

100

120

140

160

0 20 40 60 80 100 120 140 160

Departures per Hour

Arr

ival

s pe

r Hou

r

ASPM - April 2000 - Instrument Approaches

Calculated IMC Capacity

ASPM - April 2000 - Visual Approaches

Optimum Rate (ORD)

100,100

80,80

Calculated VMC Capacity

– Delay = Arrival Delay + Queuing Delay

Good weather Condition: N(0,52)

UP-Front Payment vs. Cash-Flow Royalty

Auction Proceeds could be paid out to the FAA on a monthly basis (i.e. License Royalty Fee to Reserve Arrival Time Slot)FAA could retain a % to replace ATC ticket taxAirport could use a % to replace PFC tax and invest in New Runways, Taxiways, etc.

Airline Avionics Investments Required to Increase Airport Capacity

Flight Management Systems with Required Time of Arrival CapabilitiesADS-B Cockpit Display of Traffic Information with the Capability of Providing Pilot Controlled Time-Based Separation

Airlines Could bid with Avionics Investment Promissory Notes

Airlines could Bid with Script that constituted a contract to equip their Aircraft with-in X years (i.e. ½ bid price)Accepted Airline Bid constitutes a Contract with the FAA to provide Operational Procedures that Utilize Decreased Separation Capabilities

LGA Arrival - Departure IMC

0

10

20

30

40

50

60

0 10 20 30 40 50 60

Departures per Hour

Arriv

als

per H

our

ASPM - April 2000 - Instrument ApproachesASPM - October 2000 - Instrument ApproachesCalculated IMC CapacityReduced Rate (LGA)

32,32

Dynamics of Over-Scheduling

ORD Scheduled Operations (Source: ASPM Dec 2003)

0

10

20

30

40

50

60

70

80

5 8 11 14 17 20 23Time

Total Operations Arrivals

optimum total rate

optimum arrival rate

Flight banking creates inefficient runway utilization

AA and UA Scheduled Operations at ORD grouped by 15-min epochs

(Source: BTS Dec 2003)

0

4

8

12

16

20

0 4 8 12 16 20

AA

UA

Departure

Arrival

Airline competition for market share