Bayesian Networks for Estimation
of Delay Propagation and
Cancellation in the NAS
Speaker: Chun-Hung Chen
Dept. of Systems Engineering & Operations Research
George Mason University, Fairfax, VA, USA
cchen9@gmu.edu
http://mason.gmu.edu/~cchen9
(Joint work with Ning Xu, Kathryn Laskey, George Donohue)
CENTER FOR AIR TRANSPORTATION
SYSTEMS RESEARCH
Delay Propagation
2
CA
TS
R
Motivation & Approach
• Delay/cancellation has steadily increased
• Delay at one airport has ripple effect to the whole
NAS
• Our Bayesian Network can
– Model complex, highly nonlinear, stochastic behaviors for
NAS
– Represent interactions between airports
– Capture the complex relationship between variables in NAS,
e.g.
• Show how the delays are propagated to different airports
under different weather conditions
• Show how cancellation is impacted by other variables
3
CA
TS
R
Notation & Representation
CA
TS
R
Node: a variable
of interest
Node’s name/ title
Probability (graphical)
Arc:
dependency
Names / ranges
for possible values
Mean & std dev
Probability (numerical percentage)
4
Bayesian Inference
5
CA
TS
R
Delay Propagation
6
CA
TS
R
Data for Constructing Model
ASPM
Airport
Weather
ETMS
City Pair
All Flight
By Departure
By Arrival
WindSpeed
ArrivalDemand
ArrDelay
DepDelay_
originAirport
ArrDelay_
destinationAirport
WindAngle
DepartureDemand
DepDelay
#OfFlight
#OfFlight
Weather
CanceledDep
Visibility
Quarter Hour Report (aggregate value for 15 minutes)
RunwayConfig
Primary key (index) : Date, Local hour(TimeOfDay), Qtr
AAR
ADR
7
CA
TS
R
3 months (11/2003~01/2004), 8832 instances (cases)
Each case contains an observed value for all variables
ORD: Delay & Cancellation
8
CA
TS
R
Example: Comparing Airports
VMC
ORD
LGA
9
IMC
CA
TS
R
Flight Cancellation (ORD)
Winter vs. Summer; Morning vs. Evening
10
CA
TS
R
Delay Propagation from ORD
11
CA
TS
R
Delay Propagation Into ATL
Scenario 1 - ORD Delay
Scenario 2 - LGA Delay
ORD
ORD
LGA
12
CA
TS
R
Scenario 3 - Both Delay
LGA
Decomposing Delay
CA
TS
R
Expected Value of Attributing Delays from ORD
given Gate Arrival Delay at ATL
PreGateInDla(ORD)
TurnAroundDla(ORD)
GateOutDla(ORD)
TaxiOutDla(ORD)
AirborneDla(ORD,ATL)
TaxiInDla(ATL)
Delay (minutes)
70
50
30
10
-10
0~15
15~30
30~45
45~60
60~75
75~90
90~105 105~120
>=120
ATL Gate Arrival Delay (minutes)
Given evidence for GateInDelay(ATL,ORD)
13
What If Scenarios
14
CA
TS
R
Conclusions
• Constructed stochastic model to represent
relationship for variables in NAS
• Applied model to analyze cause & effect
CA
TS
R
– Impact of weather on delay & cancellation
– Effects of season and time of day
– Propagation of delay from airport to airport
– Major contributing factors to each phase of delay
• Model is complementary with simulation
models
15
Acknowledgements
• This research has been partially supported
by NASA and NSF
• The following individuals from NASA Ames
provided valuable comments, suggestions
and support:
Drs. Harry Swenson, Karlin Roth, Larry Meyn,
Kevin James
16
CA
TS
R
Thank You
17
CA
TS
R