impact of eobt uncertainty on airport surface congestion
TRANSCRIPT
Sandeep Badrinath & Hamsa BalakrishnanMIT Department of Aeronautics & Astronautics
Emily Joback & Tom Reynolds ([email protected])MIT Lincoln Laboratory
NASA ATD-2 Industry Workshop4-5 September 2019
Impact of EOBT Uncertainty on Airport Surface Congestion Management
Sponsor: Chris Dorbian, FAA Office of Environment & Energy
EOBT Analysis - 2TGR 09/05/19
• Airport surface congestion leads to increased taxi time, fuel burn & environmental impacts
• Advanced automation systems are under development to reduce surface congestion– FAA Terminal Flight Data Manager (TFDM)– NASA Airspace Technology Demonstrator-2
(ATD-2)
Motivation
• Effectiveness of systems depend on algorithm design and accuracy of key input data, e.g., Earliest Off-Block Time (EOBT)
Need for analysis to understand relationship between automation system benefits and EOBT accuracy to inform future algorithm design and airline investments
EOBT Analysis - 3TGR 09/05/19
Motivation
Congestion management
algorithm
AirportRunway Configuration
Earliest Off Block Time (EOBT)
Target Off Block
Time (TOBT)
…Taxi-out Time Prediction
EOBT: 1205Z
EOBT: 1210Z
EOBT: 1215Z
EOBT: 1203Z
EOBT: 1220Z
Large queue
Default procedures: aircraft push when ready
Congestion management procedure: aircraft publish EOBTs
09 27
09 27
EOBT Analysis - 4TGR 09/05/19
09 27Motivation
Smaller queue
Congestion management
algorithm
AirportRunway Configuration
Earliest Off Block Time (EOBT)
Target Off Block
Time (TOBT)
…Taxi-out Time Prediction
Large queue
Default procedures: aircraft push when ready
EOBT: 1215Z
EOBT: 1220Z
Congestion management procedure: aircraft publish EOBTs
09 27
EOBT Analysis - 5TGR 09/05/19
Smaller queue
Congestion management
algorithm
AirportRunway Configuration
Earliest Off Block Time (EOBT)
Target Off Block
Time (TOBT)
…Taxi-out Time Prediction
Congestion management procedure: aircraft publish EOBTs
Motivation
Aircraft held at gate with engines off
Aircraft in large queuewith engines on
Default procedures: aircraft push when ready
09 27
09 27
EOBT Analysis - 6TGR 09/05/19
• Development and validation of queuing network models for surface operations at Charlotte (CLT)
• Evaluation of a candidate congestion management algorithm using queuing model (NASA’s ATD-2) at CLT
• Estimation of levels of EOBT uncertainty in currently reported data at Charlotte (CLT), Dallas (DFW) & Newark (EWR)
• Assessment of the impact of EOBT uncertainty on the performance of congestion management algorithm
Outline
Need for analysis to understand relationship between automation system benefits and EOBT accuracy to inform future algorithm design and airline investments
EOBT Analysis - 7TGR 09/05/19
• Need queuing model to allow with/without surface congestion management comparisons
Queuing Network Model of CLT
Charlotte airport layout Queuing network representation
36L 36C 36R
Taxi-in ramp Taxi-out ramp
Runway crossing
Runway queue
GateArrivals
Departures
EOBT Analysis - 8TGR 09/05/19
• Comparison between the model and data for a typical day (06/25/16)
Model Validation: Typical Day
Runway queue: 36C
Local time (hr) Local time (hr)Taxi
-out
runw
ay q
ueue
36C
Local time (hr)
Taxi
-out
ram
p qu
eue
Ramp queue
Taxi
-out
runw
ay q
ueue
36R Runway queue: 36R
EOBT Analysis - 9TGR 09/05/19
• Comparison between the model and data for a typical day (06/25/16)
Model Validation: Typical DayAv
erag
e ta
xi-o
ut
time
(min
)
Local time (hr)
15-min average taxi-out time
EOBT Analysis - 10TGR 09/05/19
• Error statistics on an independent test set: 7,484 departures, May/June 2016, CLT northflow
• Queuing network model can be adapted to other airports– Extended to DFW and EWR for this phase of the analysis
Model Validation: Aggregate Statistics
Statistics (min)
Gate to spot
Spot to runway Taxi-out
Mean values 9.68 10.40 20.09 Mean error -0.87 -0.58 -1.45 Mean |error| 3.08 2.7 4.35 % flights with |error| < 5 mins
82% 86% 69%
Taxi-out time (min)
Rel
ativ
e fr
eque
ncy model
data
EOBT Analysis - 11TGR 09/05/19
• Development and validation of queuing network models for surface operations at Charlotte (CLT)
• Evaluation of a candidate congestion management algorithm using queuing model (NASA’s ATD-2) at CLT
• Estimation of levels of EOBT uncertainty in currently reported data at Charlotte (CLT), Dallas (DFW) & Newark (EWR)
• Assessment of the impact of EOBT uncertainty on the performance of congestion management algorithm
Outline
Need for analysis to understand relationship between automation system benefits and EOBT accuracy to inform future algorithm design and airline investments
EOBT Analysis - 12TGR 09/05/19
Congestion Management Algorithm: Ideal Case
(a) Default scenario (no congestion management)
Take-off time
Taxi-out time
Unimpeded time Wait time
Push ready time
EOBT Analysis - 13TGR 09/05/19
Congestion Management Algorithm: Ideal Case
(a) Default scenario (no congestion management)
Take-off time
Taxi-out time
Unimpeded time Wait time
(b) Congestion management(ideal case)
Gate hold time = predicted wait
time
Taxi-out time = unimpeded time
TOBT (new gate release time)
Push ready time
TOBT = Target Off Block Time
EOBT
EOBT Analysis - 14TGR 09/05/19
• Buffer parameter accounts for errors in taxi-out time prediction, EOBT and other sources, in order to avoid losing runway utilization
• ATD-2 logic: TOBT = EOBT + max(0, Predicted wait time – Buffer)
Congestion Management Algorithm: ATD-2 logic
(a) Default scenario (no congestion management)
Push ready time Take-off time
Taxi-out time
Unimpeded time Wait time
(b) Congestion management (ATD-2 algorithm logic)
Gate hold time = predicted wait time -
buffer Unimpeded time
TOBT (new gate release time)
BufferTaxi-out time
EOBT
EOBT = Earliest Off Block TimeTOBT = Target Off Block Time
EOBT Analysis - 15TGR 09/05/19
Congestion Management Algorithm: ATD-2 logic
(a) Default scenario (no congestion management)
Push ready time Take-off time
Taxi-out time
Unimpeded time Wait time
(b) Congestion management (ATD-2 algorithm logic)
Gate hold time = predicted wait time -
buffer Unimpeded time
TOBT (new gate release time)
BufferTaxi-out time
EOBT
EOBT = Earliest Off Block TimeTOBT = Target Off Block Time
• Buffer parameter accounts for errors in taxi-out time prediction, EOBT and other sources, in order to avoid losing runway utilization
• ATD-2 logic: TOBT = EOBT + max(0, Predicted wait time – Buffer)
EOBT Analysis - 16TGR 09/05/19
• Departure metering logic tested using stochastic simulations– 6,447 departures over 15 day period at CLT
• Taxi-out time reduction depends on the choice of excess queue buffer (larger the buffer, lower the benefits)
• Optimal buffer is lowest value that ensures no loss in runway utilization
• Results with a planning horizon of 20 min
Congestion Management: Perfect EOBT information
Optimal buffer
20 minute planning horizon
Tim
e (m
in)
Buffer time (min)
EOBT Analysis - 17TGR 09/05/19
• Development and validation of queuing network models for surface operations at Charlotte (CLT)
• Evaluation of a candidate congestion management algorithm using queuing model (NASA’s ATD-2) at CLT
• Estimation of levels of EOBT uncertainty in currently reported data at Charlotte (CLT), Dallas (DFW) & Newark (EWR)
• Assessment of the impact of EOBT uncertainty on the performance of congestion management algorithm
Outline
Need for analysis to understand relationship between automation system benefits and EOBT accuracy to inform future algorithm design and airline investments
EOBT Analysis - 18TGR 09/05/19
• Many airlines publish EOBT data through TFMS feed– EOBT messages
from a major airline shown here
• EOBT error(t) = EOBT(t) – AOBT
• EOBT error varies for different look-ahead times
Evaluation of Empirical EOBT Uncertainty
EOBT error (min)
Prob
abili
ty d
ensi
ty
CLT: 12/13/2017 –02/12/2018
Lookahead time
AOBT = Actual Off Block Time
EOBT Analysis - 19TGR 09/05/19
Evaluation of Empirical EOBT Uncertainty
EOBT error (min)
Prob
abili
ty d
ensi
ty
DFW: 05/01/2018 –07/31/2018
Lookahead time
Lookahead time
EWR: 01/01/2018 –11/11/2018
• Many airlines publish EOBT data through TFMS feed– EOBT messages
from a major airline shown here
• EOBT error(t) = EOBT(t) – AOBT
• EOBT error varies for different look-ahead times
AOBT = Actual Off Block Time
EOBT Analysis - 20TGR 09/05/19
Evaluation of EOBT Uncertainty Summary Results
Stan
dard
dev
iatio
n of
EO
BT
erro
r (m
in)
Lookahead time (min)
• CLT: 12/13/2017 –02/12/2018
• DFW: 05/01/2018 –07/31/2018
• EWR: 01/01/2018 –11/11/2018
EOBT Analysis - 21TGR 09/05/19
• Development and validation of queuing network models for surface operations at Charlotte (CLT)
• Evaluation of a candidate congestion management algorithm using queuing model (NASA’s ATD-2) at CLT
• Estimation of levels of EOBT uncertainty in currently reported data at Charlotte (CLT), Dallas (DFW) & Newark (EWR)
• Assessment of the impact of EOBT uncertainty on the performance of congestion management algorithm
Outline
Need for analysis to understand relationship between automation system benefits and EOBT accuracy to inform future algorithm design and airline investments
EOBT Analysis - 22TGR 09/05/19
• EOBT uncertainty impacts congestion management because of– Reduced prediction accuracy of taxi-out times– Non-conformance to the target pushback time
• Need to increase excess queue buffer parameter to account for uncertainties and to maintain runway utilization
Congestion Management in the Presence of EOBT Uncertainty
20 minute planning horizon
Std. dev. of EOBT error (min)
Opt
imal
buf
fer t
ime
(min
)
Std. dev. of EOBT error (min)
Taxi
-out
tim
e re
duct
ion
(min
)
EOBT Analysis - 23TGR 09/05/19
Congestion Management in the Presence of EOBT Uncertainty
Ann
ual f
uel b
urn
savi
ngs
(Mkg
)
Std. dev. of EOBT error (min)
Taxi-out time & fuel benefits decrease as EOBT error increases
EOBT Analysis - 24TGR 09/05/19
• Surface congestion management automation systems will enable fuel and emissions reductions
• Analysis presented to understand relationship between automation system benefits and input data (esp. EOBT) accuracy
• Informs future algorithm design and airline investment decisions
• Recommended next steps– Extend analysis to broader range of airports and operating conditions– Analyze incentives for airlines to improve the accuracy of EOBT data– Develop and evaluate surface congestion algorithms that can
• Explicitly handle uncertainties
• Account for uncertainties in arrival times, in addition to EOBT uncertainties
Summary
EOBT Analysis - 25TGR 09/05/19
Disclaimer
DISTRIBUTION STATEMENT A. Approved for public release: distributionunlimited. This material is based upon work supported by the FederalAviation Administration under Air Force Contract No. FA8702-15-D-0001.Any opinions, findings, conclusions or recommendations expressed inthis material are those of the author(s) and do not necessarily reflect theviews of the Federal Aviation Administration.
© 2019 Massachusetts Institute of Technology.
Delivered to the U.S. Government with Unlimited Rights, as defined inDFARS Part 252.227-7013 or 7014 (Feb 2014). Notwithstanding anycopyright notice, U.S. Government rights in this work are defined byDFARS 252.227-7013 or DFARS 252.227-7014 as detailed above. Use ofthis work other than as specifically authorized by the U.S. Governmentmay violate any copyrights that exist in this work.