International
Center
for
Air
Transportation
September 25, 2003
Georg Theis, Francis Carr, Professor JP Clarke
Presented at NEXTOR/CDM Review Meeting,
Potential Approaches to
Increase Milestone Prediction Accuracy in
NAS Movements
MIT
MIT
ICAT
Agenda
y Future Research
y Outlook: Using Real-Time Ground Handling Milestones during
Turnaround to Improve the Accuracy of Pushback Time
Prediction
y Benefits of Improved Taxi-Out Time Prediction Methods
y Milestones for NAS Movements
y CDM Mission
MIT
ICAT
2
CDM Mission
GDP Projected Demand and Capacity
Departure Delays
Planned and Actual Pushback Times
Airport Acceptance Rates
Airports Configurations
y Integrated Arrival and Departure Management
❏
❏
❏
❏
❏
(CDM Summary Report, 01/08/03)
y NAS Status Information Dissemination (NASSI)
(FAA website: http://ffp1.faa.gov/tools/tools_cdm.asp)
y “Collaborative Decision Making provides Airline Operations
Centers and the FAA with real-time access to National Airspace
System status information including weather, equipment, and
delays “
MIT
ICAT
3
flight time
on blocks
touchdown
taxi-in
ground event
Models
Estimates
e.g. Idris,
Clarke et al.
Ready
for
Takeoff
Tower,
TRACON
Airlines (e.g. Lido, proprietary)
Ready for
Touchdown
Airline
Ready
for onblocks
Tower,
TRACON
How can we come up with the missing estimates?
How do estimates benefit downstream processes?
Ready for
Pushback
Airline
4
Agents should predict when they are ready to hand the aircraft over to the next agent (“handoff”):
Published Schedule
takeoff
taxi-out
flight AA 123
Milestones for NAS movements –
where are we today?
off blocks
ground event
MIT
ICAT
Research Questions
y Outlook: How can pushback prediction be improved based on
real-time ground handling data?
y To what extent does an improved taxi-out time prediction
increase the accuracy of touchdown prediction?
MIT
ICAT
5
9
8
7
6
5
4
3
2
1
0
0
6
12
Bin
18
24
30
M
e
or
Frequency
8:09
7:55
7:40
7:26
7:12
6:57
6:43
Precalculated and Actual Flight Times LH 422
23
21
19
17
15
13
11
9
PCFT
Actual Flight Times
day of operation, Aug 2002
25
7
5
3
1
Taxi Out FRABOS LH422 01AUG31AUG
Pre-calculated flight times at day of operations and actual flight times
(“Pre-calculated flight times” are calculated two hours before departure
based on planned departure/approach path, winds, etc.)
y
Frequency
Actual taxi-out times
hh:mm
y
27
Flights LH422 FRABOS for August 2002
Available Data
29
y
MIT
ICAT
6
Taxi-Out Time
31
Two Steps
y Part B: Estimation of average wait times and runway idle time
based on touchdown predictions
y Part A: Prediction of touchdown time based on the available
data
MIT
ICAT
7
Taxi-Out Time
Part A: Methodology
y Compare standard deviations of touchdown time
y Convolve “taxi out deviation distribution” and “flight time
deviation distribution”
❏ Base case:
o taxi-out prediction: “actual taxi-out time deviation” from taxi-out mean
o flight time prediction: “actual flight time deviation” from PCFT
❏ Test case:
o improved taxi-out prediction:
50% of “actual taxi-out time deviation” from taxi-out mean
o flight time prediction: actual flight time deviation” from PCFT
y Predicting touchdown based on different levels of taxi-out time
prediction accuracies
MIT
ICAT
8
Taxi-Out Time
MIT
ICAT
0
-15
0.05
0.1
0.15
0.2
0.25
0
-15
0.02
0.04
0.06
0.08
0.1
0.12
0.14
-10
-10
-5
-5
0
0
5
5
10
10
15
15
+
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0
1
2
3
4
5
6
7
8
9
10 11 12
Flight Time Prediction Deviation
(Act. FT – PCFT)
9
Taxi-Out Time
Part A: Methodology
Taxi-out Time Prediction Deviation
(Act. Taxi – Mean Taxi)
Base Case
Test Case
-5
0
5
10
Dev. Actual Taxi Out from Median of Taxi Out
-10
0:00
0:01
0:02
0:04
0:05
0:07
0:08
0:10
0:11
0:12
15
10
Taxi-Out Time
Part A: Calculations
No significant correlation between taxi-out time
prediction or flight time prediction
MIT
ICAT
Dev. Actual FT from PCFT
0
-15
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
-5
0
5
10
15
σ= 6.03 min
-10
Base Case
20
25
0
-15
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
-5
0
5
10
15
σ= 4.05 min
-10
Test Case
20
25
Part A: Convolution Results
Deviation from touchdown prediction
MIT
ICAT
11
Taxi-Out Time
Discussion Part A
y More accurate taxi-out time prediction has a considerable effect
on the accuracy of touchdown predictions
y In this example, an improvement of taxi-out time prediction by
50% leads to an improvement of touchdown prediction by
approx. 33%
MIT
ICAT
12
Taxi-Out Time
y Simulation: 25 Runs
❏ Runway only used for landings
❏ Service Time 1.5 min, i.e. capacity of 40 aircraft/hour
❏ Expected headway of landing aircraft 1.5 min. with different standard
deviations for base and test case
❏ 16 hour operation
❏ First come first serve
y Assumptions:
y Analyze average wait time and average runway idle time for
different touchdown prediction accuracies at constant expected
arrival headways
MIT
ICAT
13
Taxi-Out Time
Part B: How does improved
touchdown prediction change wait times and
runway idle time?
y
Wait time per aircraft:
y
Time when no aircraft is serviced (because no aircraft is available in queue)
Runway Idle Time:
Departure Time – Service Time – Arrival Time
Standard Deviation as calculated in Part A
-> set E(Touchdown Deviation)=0
+ [PCFT + Mean (DeviationPCFT vs. Actual Flight Time)
[Median Taxi-Out + Mean (DeviationMedian vs. Actual Taxi Out)]
E(Touchdown)=
Part B: Methodology
y
y
MIT
ICAT
14
Taxi-Out Time
Pushback
Origin D
Origin B
Pushback
Origin A
Pushback
MIT
ICAT
Origin C
0
1.5
3
4.5
Touchdown
Touchdown
Base Case: µ=0, σ=6.03
Pushback
Touchdown
Touchdown
Destination Airport
Base Case
Origin C
Pushback
0
1.5
3
4.5
Touchdown
Touchdown
Touchdown
Touchdown
Destination Airport
t/min
t/min
15
Test Case: µ=0, σ=4.05
Origin D
Pushback
Origin B
Pushback
Origin A
Pushback
Test Case
Aircraft have an expected
arrival every 1.5 minutes
Taxi-Out Time
Results
6.03 (base case)
4.05 (Test Case)
Headway Idle Time Aver. Wait Idle Time Aver. Wait
1.5
1.45% 5.72 min 0.95%
4.68 min
y Idle times can be reduced by 34%
y Wait times can be reduced by approximately one minute by
using improved taxi-out time prediction
MIT
ICAT
16
Taxi-Out Time
Headway
1.4
1.45
1.5
1.55
1.6
17
4.05 (Test Case)
3.13 (perfect taxi)
Idle Time Wait Time Idle Time Wait Time
33.66499
0.51%
32.88
0.48%
18.33
0.48%
17.65
0.46%
4.68
0.87%
4.16
0.95%
2.74
3.86%
2.44
4.10%
2.26
6.83%
1.94
6.80%
Changing the headway
gives consistent results
6.03 (base case)
Idle Time Wait Time
0.87%
33.68
0.93%
18.44
1.45%
5.72
4.48%
3.9
7.29%
2.7
MIT
ICAT
Taxi-Out Time
1
0.00%
10
100
MIT
ICAT
Aver. wait time per a/c in min.
4.00%
Base Case
2.00%
Test Case
8.00%
Perfect Taxi Out Pred.
6.00%
σ=6.03
σ=4.05
σ=3.13
Runway Idle Time Share
E (Headway)=
service time=1.5 min
Managed Arrival Reservoir
Results
10.00%
18
Taxi-Out Time
Implications
❏ At a value of 30$ per hour and 100 pax/aircraft, reduction of another
$11.7 Mio p.a.
o 10.7 delay hours/day*30*100=$32100/day
o $32100*365 days= $11.7 Mio p.a.
y Passenger delay costs:
❏ At $3000/block hour, reduction of $ Mio 11.7 p.a.
o 40aircraft/hour*1min*16 hours=10.7 delay hours/day
o 10.7 delay hours/day*365 days*$3000=$ Mio 11.7 p.a.
y Reduction in holding patterns:
y Wait times in holding patterns can be reduced and runway
utilization rates improved
19
y More accurate taxi-out time can improve the prediction accuracy
of touchdown
MIT
ICAT
Taxi-Out Time
Limitations/Next Steps
❏ Verify results with larger sample and more simulation runs
❏ Allow for dual use runway and different aircraft sizes
❏ Allow changes in headway over time
y Possible Next Steps
❏ Small sample
❏ Assumption was made that pushback at origin can be moved arbitrarily
to assure an arrival at a specific expected time
❏ Assumption that pushback time setting has no effect on taxi/flight time or
en-route constraints
❏ Headway is fixed
y Limitations
MIT
ICAT
20
Taxi-Out Time
Pushback Prediction
-> We need an accurate pushback time prediction: Ongoing
Research of Francis Carr/Georg Theis/Professors Clarke, Feron
y How can we already predict the downstream touchdown during
turnaround?
y Controllers need to know how many aircraft are about to push
21
Outlook: How can ground handling data be used to
improve the accuracy of pushback time prediction?
y Improved Taxi-Out Time Prediction only helps after Pushback
MIT
ICAT
Aircraft Acceptance
Towing
Un-/Loading
Fueling
Cleaning
Catering
Not Ready Message
Pushback
Passenger Buses
Boarding
Deicing
❑ For all processes, beginning and end are reported real-time
❑ For every involved vehicle, the system gets the messages “at aircraft”, “begin service”, “end service” 22
Aircraft
Positioning
Gate Processes
Crew Processes
Passenger Buses
Deboarding
Core processes
Standard processes
Exceptional processes
Ground Handling Data to our knowledge
currently only available to Lufthansa Airlines
ALLEGRO measures crew on position,
beginning and end of all main turnaround processes
MIT
ICAT
Aircraft Acceptance
Towing
Un-/Loading
Fueling
Cleaning
Catering
Not Ready Message
Pushback
Passenger Buses
Boarding
Deicing
Standard processes
Exceptional processes
Source (translated): Theis, G. (2002) Telematik Anwendungen im Luftverkehr,
Internationales Verkehrswesen, 54(5), 225-228.
See also: http://www.fraport.com/online/general/en/download/presentation_220802.pdf, 12-13
Aircraft
Positioning
Gate Processes
Crew Processes
Passenger Buses
Deboarding
Core processes
Used Data
ALLEGRO measures crew on position,
beginning and end of all main turnaround processes
MIT
ICAT
23
Methodology
❏ “On block based”:
Sched.TDep= max (Sched.TDep, (on block + Minimum Ground Time))
❏ Age based: use elapsed ground time to improve updates
❏ Status based: update pushback prediction after end of major processes
y Different methods
y With these causes being eliminated,
“ready for pushback = actual pushback” from a
ground handling perspective
❏ GDP
❏ Technical Failure
❏ Weather
y Filtered out exogenous influences
MIT
ICAT
24
Preliminary Results
Pushback Prediction
❏ Subsequent processes catch up for late previous processes, i.e. basing
the prediction on planned process times reduces the estimate’s quality
❏ Decision Support Tools for ground-air handoffs will have to take into
account intrinsic stochasticity of pushback estimates
y However, uncertainty in airline turn process has significant
lower bound
y Status based predictions improve the quality of pushback
estimates
MIT
ICAT
25
Future Research
y Estimate Potential Benefits
y Make Models Responsive to Dynamic Stream of Data Updates
❏ Taxi Out Time Model (Idris, Clarke, et al.)
❏ Pushback Time Prediction Model
y Check what data sources are available to use for
MIT
ICAT
26