Strategic Flight Cancellation under Ground Delay Program ... · © 2019 The MITRE Corporation. All rights reserved. Approved for Public Release 19-0503. Distribution Unlimited | 1

© 2019 The MITRE Corporation. All rights reserved. Approved for Public Release 19-0503. Distribution Unlimited

| 1 |

Christine TaylorShin-Lai (Alex) TienErik VargoCraig Wanke

The MITRE CorporationMcLean, VA

Strategic Flight Cancellation under Ground Delay Program Uncertainty

13 th USA/Europe ATM R&D SeminarVienna, Austria

17-21 June, 2019

| 2 |

©2019 The MITRE Corporation. All rights reserved. Approved for Public Release 19-0503. Distribution Unlimited

Motivation: Strategic Planning from an FOC Perspective

FOC Dispatcher

Significant AAR reduction for a sustained periodLikely Low Rate GDP

Confidence: High

Strategically Cancel Flights

Early Cancellation allows for best resource management and minimal passenger disruption

| 3 |



FOC Dispatcher

Moderate to Significant AAR reduction Potential GDP

Confidence: Medium


Risk unnecessary disruption

Wait and See

| 4 |



FOC Dispatcher

Wait and See

Significant AAR reduction for a sustained periodLikely Low Rate GDP

Confidence: High

No Notice Cancellations

Large Delays

Crew Time Outs

Newsworthy Passenger Disruptions

Moderate to Significant AAR reduction Potential GDP

Confidence: Medium


Risk unnecessary disruption

| 5 |



FOC Dispatcher

GDP: Time/RatesProbability: 25%



Strategically cancel a few flightsLook for opportunities to recover

Wait and See

Strategically cancel later flights

Which (if any) flights should be cancelled now?

| 6 |


Adaptive Planning Framework

Current decision

node

TimeNow

Recommended action given current forecast

Adaptive strategy to optimize performance under predicted future scenarios

Future decision points

Scen. 1, prob.

Scen. 2, prob.

Scen. 3, prob.

Scen. 4, prob.

Scen. 5, prob.

Balancing Cost and Risk• Earlier decisions are less costly than later decisions• Later decisions are made under greater certainty• Some decisions are not available later• Some decisions can be changed (at a cost); others cannot

Incurred Projected

| 7 |


Generating Forecast Tree

Ensemble Weather Forecast

Probabilistic AAR Forecast Translation

GDP Emulator

GDP Scenario

Prediction

Learned Forecast Similarity

Assigned Arrival Slots

Forecast Horizon

10

0 E

nse

mb

le M

emb

ers

Current decision

node

TimeNow

Scen. 1, prob.

Scen. 2, prob.

Scen. 3, prob.

Scen. 4, prob.

Scen. 5, prob.

AAR Scenario Tree

…………

| 8 |


Strategic Flight Cancellation Problem

Current decision

node

TimeNow

Recommended action given current forecast

Adaptive strategy to optimize performance under predicted future scenarios

Future decision points

GDP 1, prob.

GDP 2, prob.

GDP 3, prob.

GDP 4, prob.

GDP 5, prob.

Cost 1

Cost 2

Cost 3

Cost 4

Cost 5

Plan Objective

Cancellation Costs

Legacy Narrowbodies: Regional Jets

4.7:1Delay Costs

Assignment costs represent passenger delay

| 9 |


Case StudyLGA GDP 13 November 2018

22Z

Sin

gle

De

cisi

on

Po

int

(SD

P)

Ad

apti

ve P

lan

nin

g Fr

amew

ork

(A

PF)

Forecast Issued

Single decision

node

GDP 1, prob.

GDP 2, prob.

GDP 3, prob.

GDP 4, prob.

GDP 5, prob.

Cost 1

Cost 2

Cost 3

Cost 4

Cost 5

Min. Expected Cost

GDP 1, prob.

GDP 2, prob.

GDP 3, prob.

GDP 4, prob.

GDP 5, prob.

Cost 1

Cost 2

Cost 3

Cost 4

Cost 5

Min. Expected CostCurrent

decision node

| 10 |


Case StudyLGA GDP 13 November 2018

22Z 02Z 09Z 11Z … …

Time GDP is Issued

Actual GDP

Sin

gle

De

cisi

on

Po

int

(SD

P)

Ad

apti

ve P

lan

nin

g Fr

amew

ork

(A

PF)

Optimize SDP Strategy



Optimize SDP Strategy Assuming

Prior Actions


Prior Actions



Prior Actions

Optimize APF Strategy



Optimize APF Strategy Assuming

Prior Actions


Prior Actions



Prior Actions

Forecast Issued

Forecast Issued

Forecast Issued

| 11 |


Scen ProbTimeline (Z)

11 12 13 14 15 16 17 18 19 20 21 22 23 24 01 02 03

S1 33%

S2 4%

S3 16%

S4 21%

S5 7%

S6 7%

S7 12%

13 Nov Forecasted GDPs:Prior-day planning (22Z)

GDP Hourly Rate

30 34 36 38GDP

None

22Z 02Z 09Z

S1

S2

S3

S4

S5

S6

S7

GDP = 61%

| 12 |


Individual Scenario Optimization

| 13 |


GDP

None

Adaptive Planning Framework vs. Single Decision Point Solutions at 22Z

| 14 |


13 Nov Forecasted GDPs:Prior-day planning (02Z)

GDP Hourly Rate

30 34 36 38GDP

None

02Z 09Z 12Z

S1

S2

S3

S4

S5

S6

S7


11 12 13 14 15 16 17 18 19 20 21 22 23 24 01 02 03

S1 73%

S2 7%

S3 4%

S4 2%

S5 2%

S6 11%

S7 1%

GDP = 85%

| 15 |


GDP

None


| 16 |


13 Nov Forecasted GDPs:Day-of Planning (09Z)

GDP Hourly Rate

30 34 36 38GDP

None

09Z 12Z 15Z

S1

S2

S3

S4

S5

S6

S7


11 12 13 14 15 16 17 18 19 20 21 22 23 24 01 02 03

S1 11%

S2 1%

S3 1%

S4 4%

S5 10%

S6 11%

S7 62%

GDP = 37%

| 17 |


GDP

None


| 18 |


GDP

None

APF vs SDP solutions @ 09Z with prior actions

| 19 |


Incurred verses Projected Cost

| 20 |


Conclusions

▪ APF permits explicit optimization of future decisions, highlighting actions that should be taken now

– Aligns decisions with information uncertainty

– Directly captures risk of inaction with cost of inappropriate response

▪ Key to successful adaptive planning is representing the forecast tree

– Similarity must reflect appropriate features of forecast evolution

– Scenarios should span the space of planning outcomes

▪ What does a validation plan look like?

– Beyond forecast tree representation (and accurately capturing costs), how do we validate probabilistic planning decisions?

Documents

Strategic Flight Cancellation under Ground Delay Program ... · © 2019 The MITRE Corporation. All rights reserved. Approved for Public Release 19-0503. Distribution Unlimited | 1