© 2015 The MITRE Corporation. All rights reserved.Joint Copyright with Sandip Roy, WSU
Approved for Public Release; Distribution Unlimited. 15-1785
Designing Traffic Flow Management Strategies Under Uncertainty
Dr. Christine Taylor
Tudor Masek
Dr. Craig Wanke
The MITRE Corporation
Center for Advanced Aviation System
Development (CAASD)
Prof. Sandip Roy
Washington State University
Department of Electrical Engineering
11th USA/Europe ATM R&D Seminar
Lisbon, Portugal23-26 June, 2015
| 2 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Strategic TFM Planning for Weather
12Z 14Z 16Z
Probabilistic forecasts identify regions of potential weather activity
What is the range and likelihood of different weather scenarios occurring?
What are the potential scenarios of ATM impacts?
What options are available to mitigate congestion and when do we have to act?
Short Range Ensemble Forecast (SREF) for 28 July, 2014
| 3 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Leveraging Weather Forecast Scenarios for TFM
12Z
14Z
16Z
Weather traverses just north of the NY Metro
12Z
14Z
16Z
12Z
14Z
16Z
Weather traverses the NY Metro
Weather traverses far north of the NY Metro
| 4 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Forecast Uncertainty Complicates Planning
16Z
Weather traverses just north of the NY Metro
16Z 16Z
Weather traverses the NY Metro
Weather traverses far north of the NY Metro
www.publicdomainpictures.net
TMI Plan for S1
TMI Plan for S3
TMI Plan for all Scenarios
TMI Plan for S2
S1Probability Low (2/21)
Impact HighS2
Probability Low (2/21)
Impact ModerateS3
Probability High (17/21)
Impact Low
Current TMI Plan & Advisories
Individual Scenario OptimizationDesigns the best strategy for each scenario
Robust OptimizationDesigns the best strategy considering all scenarios
Adaptive Optimization
Designs the best strategy considering scenario evolution and opportunity to adapt
TMI = Traffic Management Initiative
| 5 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
The Adaptive Planning Framework
Provides strategic TFM decision support under forecast uncertainty
– Quantifies when, where, and how forecasts diverge
– Recommends executable plans (actions) and contingency plans (advisories)
– Capture trade-offs of acting now vs. waiting
– Accounts for the ability to adapt to new information
Wide range of literature on look-ahead decision-making (sequential decision-making via dynamic programming, model predictive control, etc.)
New features of this approach are:
– Uncertainties represented as spatio-temporal (trajectory) scenarios
– Realistic modeling of Traffic Management Initiatives (TMIs)
– Heuristic optimization enables large-scale simulation in the loop
| 6 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Develop Scenarios of Weather-Impact Evolution
Identify critical features
Measure Forecast
SimilaritiesTranslate to congestion
forecast
Forecast Ensemble Weather Impact Ensemble
Scenarios of Weather and
Capacity Impact
Demand forecast
Fast-time Traffic
SimulationImpact S1: p =2/21
S2: p =2/21
S3: p= 17/21
12z-14z
| 7 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Construct Adaptive Planning Framework
Forecast Look-ahead12Z 14Z 16Z
All forecasts
S1 Cluster
S2 & S3 Cluster
S1 Cluster
S3 Cluster
S2 ClusterCurrent Decision
Contingency Plan
Contingency Plan
Contingency Plan
Contingency Plan
Contingency Plan
Strategy 1
Strategy 2
Strategy 3
S1: p =2/21
S2: p =2/21
S3: p= 17/21
| 8 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Defining the TMI Design Space
FCAA08
BOSGDP
JFKGDP
EWR GDP
LGA GDP
GDP Parameters
Rate (#/hour)Start time (UTC)Duration (hours)Tier (distance)
AFP Parameters
Rate (#/hour)Start time (UTC)Duration (hours)
TMIs
Decision Periods
12 14 16 18 20
BOS EWR JFK LGA FCA
Design Space
| 9 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Total Ground Delay
Total Sector Delay
Total Arrival Schedule Delay
Other Delay Statistics
Throughput (Airports, Airspace)
Schedule Integrity
Surface Congestion
Expected value
Receding time horizon
Minimize maximum impact
Satisfy a goal/constraint
Develop robust (minimum variance) solutions
Quantifying Strategy Performance
Strategy Objective
Function of Cost and Scenario Probability
Cost
Measures Effectiveness of Strategy for Scenario
Strategy Objective
𝑆𝑂 =
𝑡
𝛼𝑡
𝑘
𝑝𝑘(𝑡) C𝑘(𝑡)
Cost
C𝑘 =
𝑡
𝐺𝐷𝑘(𝑡)+ 2∗𝑆𝐷𝑘(𝑡)
Cost 1
Cost 2
Cost 3
Strategy 1
Strategy 2
Strategy 3
S1: p =2/21
S2: p =2/21
S3: p= 17/21
| 10 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Design Space
Genetic Algorithm Overview
Genetic Algorithm
Construct Genome
Strategy Objective
Evaluate Fitness
BOS
EWR
FCA
JFK
LGA
BOS
EWR
FCA
JFK
LGA
Fitness Fitness
BOS
EWR
FCA
JFK
LGA
Fitness
FCA
Process continues until termination criteria occurs
Genetic Algorithms (GAs) are a class of heuristic search methods that mimic biological evolution
| 11 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Design Space
Optimizing Adaptive Strategies
Genetic Algorithm
Construct Genome
Strategy Objective
Evaluate Fitness
Flow Contingency Management (FCM)
Strategy Objective
𝑆𝑂 =
𝑡
𝛼𝑡
𝑘
𝑝𝑘(𝑡) C𝑘(𝑡)
Cost
C𝑘 =
𝑡
𝐺𝐷𝑘(𝑡)+ 2∗𝑆𝐷𝑘(𝑡)
Scenarios of Weather and
Capacity Impact
Demand forecast
Fast-time Traffic
Simulation
TMI Strategies
Costs
Cost 1
Cost 2
Cost 3
Strategy 1
Strategy 2
Strategy 3
S1: p =2/21
S2: p =2/21
S3: p= 17/21
| 12 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
WI Scenario TMIs
Planning Horizon (UTC)
12 14 16 18 20
S1
BOS 55 55 55 55
EWR 40
JFK 35 35
LGA 30 30 30
FCA 70
S2
BOS 55 55
EWR 25 25
JFK 50
LGA
FCA 70
S3
BOS
EWR 25 25 25
JFK
LGA
FCA
Individual Scenario Optimization
Genetic Algorithm
C𝑘 =
𝑡
𝐺𝐷𝑘(𝑡)+ 2∗𝑆𝐷𝑘(𝑡)
Genetic Algorithm
Genetic Algorithm
Requires Additional Integration
Cost 1
Cost 2
Cost 3
Strategy 1
Strategy 2
Strategy 3
S1: p =2/21
S2: p =2/21
S3: p= 17/21
| 13 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Robust Scenario Optimization
Genetic Algorithm
C𝑘 =
𝑡
𝐺𝐷𝑘(𝑡)+ 2∗𝑆𝐷𝑘(𝑡)
Cost 1
Cost 2
Cost 3
Robust Scenario
S1: p =2/21
S2: p =2/21
S3: p= 17/21
𝑆𝑂 =
𝑡
𝛼𝑡
𝑘
𝑝𝑘(𝑡) C𝑘(𝑡)
Strategy Objective
WI Scenario TMIs
Planning Horizon (UTC)
12 14 16 18 20
S1
BOS 55 55 55 55
EWR 40
JFK 35 35
LGA 30 30 30
FCA 70
S2
BOS 55 55
EWR 25 25
JFK 50
LGA
FCA 70
S3
BOS
EWR 25 25 25
JFK
LGA
FCA
Individual Scenario Optimization
WI Scenario TMIs
Decision Periods
12 14 16 18 20
All
BOS 40
EWR 25 25 25
JFK
LGA
FCA
| 14 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
WI Scenario TMIs
Decision Periods
12 14 16 18 20
S1
BOS 40 40 40 40
EWR 25 25 25 25
JFK 50 50 50
LGA 30 30 30
FCA
S2
BOS 40 55 55 55
EWR 40 40
JFK 35 35
LGA 20 20 20
FCA 90 90
S3
BOS 40
EWR 40
JFK 35 35
LGA
FCA 90
Adaptive Scenario Optimization
S1: p =2/21
S2: p =2/21
S3: p= 17/21
Genetic Algorithm
Strategy 1
Strategy 2
Strategy 3
C𝑘 =
𝑡
𝐺𝐷𝑘(𝑡)+ 2∗𝑆𝐷𝑘(𝑡)
Cost 1
Cost 2
Cost 3
𝑆𝑂 =
𝑡
𝛼𝑡
𝑘
𝑝𝑘(𝑡) C𝑘(𝑡)
Strategy Objective
| 15 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Comparing Strategy Performance
WI Scenarios Probability Robust Adaptive
S1 2/21 57757 45274
S2 2/21 13876 13201
S3 17/21 4305 4450
Strategy Objective
10307 9172
12 Z Decision
Robust EWR 25
Adaptive BOS 40
| 16 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
Summary
Adaptive Planning Framework captures the trade-off between waiting and action under forecast uncertainty
– Promotes incremental decision making by formalizing the ability to adapt
– Identifies actions that may be critical for managing high-impact futures
Continuing research focus
– How do different costs and strategy objectives affect solutions?
– How does forecast uncertainty affect hedging options?
– What constraints on solution adaptability are required to ensure operational relevance?
– Can model (network) structure be exploited to improve GA performance?
| 17 |
Approved for Public Release; Distribution Unlimited. 15-1785© 2015 The MITRE Corporation. All rights reserved. Joint Copyright with Sandip Roy, WSU
NOTICE
This work was produced for the U.S. Government under Contract DTFAWA-10-C-00080 and is subject to Federal Aviation Administration Acquisition Management
System Clause 3.5-13, Rights In Data-General, Alt. III and Alt. IV (Oct. 1996).
The contents of this document reflect the views of the author and The MITRE Corporation and do not necessarily reflect the views of the FAA or the DOT.
Neither the Federal Aviation Administration nor the Department of Transportation makes any warranty or guarantee, expressed or implied, concerning the content
or accuracy of these views.
2015 The MITRE Corporation. All Rights Reserved.
Joint copyright held by
Sandip Roy, Washington State University