Transcript LMINet2

LMINET2: An Enhanced LMINET
Dou Long, Shahab Hasan
[email protected]
December 10, 2008
Old LMINET
• Refers to a suite of models
–
–
–
–
Separate commercial and GA traffic schedule forecast models
NAS-wide operations/delay model
NAS-wide projected throughput schedule construction model
Utilities to process OAG, ETMS, ASQP data sources
• Flight delay estimation, at 110 large airports
– Based on the solution of dynamic queuing equations at 110 airports
from demands and capacities
– Outputs queuing delays (akin to measuring against unimpeded)
• Constrained flight schedule construction, all flight
– Can use delay tolerance-based rules or demand/capacity ratio-based
rules for airport capacity constraints
– Uses sector loading for en route constraints
– Runs assuming universally good weather conditions across NAS to
reflect general airline scheduling practice
PAGE 2
Motivation for LMINET2
• Although LMINET is capable of estimating flight delays in any
weather conditions, its modeling of NAS operations in bad
weather can be improved. We need to have new capabilities to
reflect the disruption of flight schedules:
– Delay propagation
– Flight cancellation (not the same as flight trimming)
– Active air traffic management (ATM) measures such as the Ground
Delay Program (GDP)
• Some ATM technologies/strategies are designed to help the traffic
operations in both good and bad weather, or in bad weather only
PAGE 3
Triad of Models and Entities in LMINET2
• The airport queuing delay model is replaced by a triad of models:
demand, flight, aircraft
Flight
Arrival
Delay
Queuing
Delay
Airport Queuing
Delay Model
Airport
Demand
Aircraft Connection
and Turnaround
Model
Flight Schedule
Delay &
Cancellation Model
Flight
Departure
Delay
Demand
Flight
Aircraft
Aggregate number of
flights
Scheduled service (O&D,
time, equipment)
For flight connection
and delay propagation
For queuing delay
calculations
Schedule delays depends
on queuing delay, schedule
pad, and schedule delay
PAGE 4
Modeling Flight Delays at Arrival Gate
and Schedule Pad Estimation
•
Flight schedule delay (gate arrival delay)
•
–
Block Delay = Taxi-out Delay + En Route Delay + Taxi-in Delay – Schedule pad, or
–
Arrival Gate Delay = Departure gate delay + Taxi-out Delay + En Route Delay + Taxiin Delay – Schedule pad
Schedule pad estimation
En
Route
(1.27)
•
Dept
Gate
Taxiout
(10.98)
(4.68)
Block
Taxi-in
(-2.19)
(1.71)
Arrival
Gate
(7.55)
Pad is 9.85 min (1st formula), or 11.09 min (2nd formula)
–
–
Their difference is caused by the rounding error in the reported data base
Data source: 2005 ASPM, all flights including negative delays
PAGE 5
Modeling Flight Connection
• Flight connection model is needed for delay propagation
and flight cancellation modules
• Only the flights with the same seat size categories can be
connected
– The carrier flag is ignored because the model is envisioned to be used
mostly for studies of future traffic when the carrier is the hardest to predict
in a flight schedule
• Window of flight connection construction
– Estimated by the scheduled arrival and departure times (not the real
operation times)
– Data sources: ASQP (tail #), and OAG (seat size) of June 2005
PAGE 6
Delay Propagation Model and
Its Validation
Model for the departure gate delay (against schedule)
= max(arrival delay + minimum ground turn time – scheduled ground turn time, 0) +
adjustment factor caused by other reasons
400
350
Actual departure delay
•
300
250
200
150
100
50
0
0
50
100
150
200
250
300
350
400
Predicted departure delay
–
Y = α + βx
–
α =7.64 , β =0.966, R2 = 0.9368
–
α =0,
β=1.05,
R2 = 0.9748
PAGE 7
Operational Flight Cancellation Module
• The following flights are cancelled
– Due to congestion based on the queuing delay
– Due to schedule delay
– Due to connectivity
• The next leg, if it exists, of a cancelled flight is also cancelled
• Flights can also be cancelled by the Ground Delay
Program logic (discussed on next slides)
PAGE 8
GDP Logic
• A proactive ATM program to reduce flight delay and congestion
when the capacity of the destination airport is reduced due to
weather by holding flights at their departure airports
• While running the normal delay model, concurrently check the
future capacity/demand imbalance at each of the 310 airports
starting 2 hours ahead till the end of day
• The acceptance rate at the destination airport can be taken
from an input file, or can be generated based on the weather
condition
• If not departed, the departure times of the arriving flights are
delayed to the next epoch. FIFO scheme used for multi-epoch
delay.
• Cancel flights if they are expected to experience extreme
delays
PAGE 9
Expanded Airport Coverage in LMINET2
• Airports with queuing delays: 310
– 110 with FAA capacity models
– 200 LMI-developed models
• Schedule delays: all commercial airports, ~ 450
• All airports: contribute demand to the 310 airports and to their delays
16,000,000
14,000,000
14,000,000
12,000,000
12,000,000
Operations per Year
Operations per Year
Air Taxi
Air Carrier
16,000,000
10,000,000
10,000,000
8,000,000
AC out
8,000,000
AC in
6,000,000
6,000,000
4,000,000
4,000,000
2,000,000
2,000,000
AT ou
AT in
0
0
Hubs 31
FACT1 BM 35
FACT2 56
ASPM 77
ATO-F/LMI 110
LMI 310
Hubs 31
FACT1 BM 35
FACT2 56
ASPM 77
P A G E 10
ATO-F/LMI 110
LMI 310
Model Parameter Calibration
• Default setting based on the system averages
• Airport specific tuning at a small set of airports
– Schedule delay pad at the arrival airport
– Non-congestion related departure delay adjustment
• They are all interconnected; there is no single parameter
responsible for one statistic
• The model output is most sensitive to airport capacity and
weather inputs
P A G E 11
Summary of Model Validation
•
We are satisfied overall for a national model
– By the delay/cancellation statistics comparison
– Because of our queuing theoretic and modular approach
•
The errors are contributed mostly by the capacity models at a few airports
– The model assumes the theoretical capacity while some airports are specified
by operational capacities.
– The capacity models assume one set of curves for each meteorological
condition. Airports may have multiple curves under weather due to different
runway configurations used, which can also cause the inconsistency of in the
GDP program.
•
Some errors are expected
– Used published commercial schedule and generated GA schedule instead of
real schedule
– It does not considered the carrier flag in flight connection for delay propagation
and flight cancellation
• Kept it this way because it is impossible to specify it in the studies of future traffic
scenarios
P A G E 12
Benefits of LMINET2:
More Realistic Setting & Richer Statistics
• It captures the delay absorption, propagation, cancellation, and
ground control
– Instead of a giant airport delay calculator, it now tracks the delays of
each individual flight
– It is especially needed in modeling NAS in bad weather
• It yields better delay estimates, even for the queuing delays
offered by the old LMINET, because of more proper accounting of
schedule disruption
• It generates a richer set of statistics in addition to queuing delays:
–
–
–
–
–
Arrival/departure gate delay
Arrival/departure gate on-time percentage (if arrival delay > 15 min)
Taxi-out delay
Arrival/departure flight cancellation statistics
These metrics provide a better representation of the current NAS
operations, and for calculating stakeholder metrics
P A G E 13
Running LMINET2
• Fast turn around
– Unlike the old LMINET for delay estimation, the computer
work load is a function of congestion
– It takes a few minutes for one day of traffic
• Inputs
– Flight schedule
– Airport capacity
– Airport weather
P A G E 14
The Future of LMINET2
• The model is ready to be used
– Better information on some parameters would be helpful
• Fine-tuning the parameters
– Will not yield a significantly better model
– But will improve the modeling at isolated areas or metrics
• It still lacks an airspace delay module to claim to be
a complete NAS operations model
• The projected throughput schedule construction is
unchanged
– It is run under universally good weather
P A G E 15