Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Prediction of passenger boarding progress using
neural network approach
Michael Schultz and Stefan Reitmann German Aerospace Center (DLR e.V.)
Institute of Flight Guidance
• Background
• Boarding model
• Long short-term memory model
• Results
• Outlook
Outline
carmodel2011.blogspot.com
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 2
• International Civil Aviation Organization (ICAO), Aviation System Block
Upgrades (ASBU) Timeline to implement efficient flight paths
• Trajectory Based Operations (TBO)
• Global ATM Operational Concept (Doc. 9854)
• Flight & Flow Information for a Collaborative Environment (Doc. 9965)
• Need for a System Wide Information Management (SWIM, Doc. 10039)
• Aircraft trajectory – not only flight path
• Milestone concept available - A-CDM
• 4D ground trajectory – including ground operations (turnaround)
• Critical path
• 70€ per delay minute
• Boarding is always on critical path (particularly important for flights
with high number of rotations, short haul)
Background
aircraft trajectory, ground operations, boarding
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 3
Eurocontrol, CODA
Background
aircraft trajectory, ground operations, boarding
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 4
-10
-5
0
5
10
15
20
25
200
82
00
92
01
02
01
12
01
22
01
32
01
42
01
5
200
82
00
92
01
02
01
12
01
22
01
32
01
42
01
5
200
82
00
92
01
02
01
12
01
22
01
32
01
42
01
5
200
82
00
92
01
02
01
12
01
22
01
32
01
42
01
5
200
82
00
92
01
02
01
12
01
22
01
32
01
42
01
5
departure time taxi-out phase flight phase taxi-in phase arrival time
tim
e (
min
)
range (80th-20th percentile) standard deviation
Variability on intra-European flights (2008-2015)
H. Fricke, M. Schultz, Delay impacts onto turnaround performance, ATM Seminar, 2009
Background
aircraft trajectory, ground operations, boarding
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 5
Ready for boarding
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 6
Data-based vs. model-based approaches
input data, models, new concepts
0
10
20
30
40
50
0 50 100 150 200
ab
so
lute
fre
qu
en
cy
passengers
fast medium slow
0
5
10
15
20
0 50 100 150 200
bo
ard
ing
tim
e (
min
)
passengers
M. Schultz, Field Trial Measurements to Validate a Stochastic Aircraft Boarding Model, Aerospace 2018, 5(1), 27 M. Schultz, Dynamic change of aircraft seat condition for fast boarding, J. of Transp. Res. Part C 2017 85:131-147
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 7
Model - A320, 29 rows, 174 seats
• grid-based, stochastic processes (ASEP), fast implementation
• Operational aspects, e.g. seat load, arrival rates, boarding strategy
• Individual passenger behaviour, e.g. speed
Aircraft boarding
stochastic, individual-based model
M. Schultz, Implementation and application of a stochastic aircraft boarding model, J. of Transp. Res. Part C 2018 90:334-349
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 8
Reliable simulation environment
Aircraft boarding
stochastic, individual-based model
M. Schultz, Implementation and application of a stochastic aircraft boarding model, J. of Transp. Res. Part C 2018 90:334-349
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 9
Implementation
boarding strategies and operational constraints
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 10
Pax could disturb the
whole boarding progress
(connection flights, pax
handling in terminal -
security)
Problem to solve
boarding is owned by the passenger
Different kind of seat
occupation pattern
demands for a specific
amount of individual
movements
1 3 5 7 9 11 13 15
0
10
20
arrival (pax per minute)
boarding time (min)
Scenario A (fast)
Scenario B (medium)
Scenario C (slow)
0
10
20
30
40
0-5 10-15 20-25 30-35 40-45
probability (%)
time (s)
triangular distribution (old model)
Weibull distribution
field measurement
Distribution and amount of
hand luggage pieces
results in blocked aisle
when storing into the
overhead compartment
M. Schultz, Field Trial Measurements to Validate a Stochastic Aircraft Boarding Model, Aerospace 2018, 5(1), 27
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 11
How to predict passenger-controlled process?
boarding progress
0 200 400 600 800 1000 1200
0
1boarding time (s)
1 -
se
at
loa
d
random boarding
outside-in boarding
0
500
Pair
cra
ft
0
600
0 200 400 600 800 1000 1200
DP
air
cra
ft
boarding time (s)
random boarding
outside-in boarding
0
360
se
qu
en
ce
eva
lua
tio
nk
M. Schultz, A metric for the real-time evaluation of the aircraft boarding progress, J. of Transp. Res. Part C 2018 86:467-487
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 12
• System state of seat rows and cabin
(using future cabin management system and sensors)
Evaluation of seat status
complexity metric
step 0
aisle window
n
n
realized movements
expected movements free seat
occupied seat (from step 2) current seat
occupied seat (from step 1)
2
2
6
2
5
5
step 2
1
1
1
step 1
3
6
7
11
9
14
step 3
1
4
1
9
4
9
10.5
8.0
3.5
8.3
M. Schultz, A metric for the real-time evaluation of the aircraft boarding progress, J. of Transp. Res. Part C 2018 86:467-487
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 13
Complexity metric
progress prediction 0 200 400 600 800 1000 1200
0
1boarding time (s)
1 -
se
at
loa
d
random boarding
outside-in boarding
0
500
Pair
cra
ft
0
600
0 200 400 600 800 1000 1200
DP
air
cra
ft
boarding time (s)
random boarding
outside-in boarding
0
360
se
qu
en
ce
eva
lua
tio
nk
0 200 400 600 800 1000 1200
0
1boarding time (s)
1 -
se
at
loa
d
random boarding
outside-in boarding
0
500
Pair
cra
ft
0
600
0 200 400 600 800 1000 1200
DP
air
cra
ft
boarding time (s)
random boarding
outside-in boarding
0
360
se
qu
en
ce
eva
lua
tio
nk
slow boarding
fast boarding
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 14
• Artificial Neural Networks (ANN) are characterized by interconnected neurons
• Neurons structured in different layers, connection depends on computed
weights, result in blocking or passing of information
Machine learning approach
artificial neural networks
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 15
• Training neural networks demands for error estimation (backpropagation)
Machine learning approach
recurrent neural network
input layer hidden layers output layer information loop
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 16
• Information to be passed from one step of the network to the next
• Long-term dependency problem (need for more context)
Machine learning approach
short-term memory
http://colah.github.io/posts/2015-08-Understanding-LSTMs/
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 17
• Repeating module in a RNN and LSTM
Machine learning approach
long short-term memory (LSTM)
forget gate
input gate
(store information)
update cell state output gate
http://colah.github.io/posts/2015-08-Understanding-LSTMs/
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 18
• Implementation in Python 3.6, TensorFlow (backend), Keras (frontend)
• Basic model: core layers (Dense, Dropout), recurrent layers (LSTM)
• Boarding model: concatenation of 1 - 3 separate LSTM branches
Machine learning approach
LSTM implementation
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 19
Machine learning approach
LSTM implementation
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 20
Input data – complexity measures
boarding simulation to train and evaluate
M. Schultz and S. Reitmann, Machine learning approach to predict aircraft boarding, J. of Transp. Res. Part C 2018 (review)
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 21
Prediction of boarding progress
M. Schultz and S. Reitmann, Machine learning approach to predict aircraft boarding, J. of Transp. Res. Part C 2018 (review)
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 22
Outlook
handle input data - smooth
0 250 500 750 1000 1250 1500 1750
Timesteps t
0
50
100
150
200
250
300
k
S. Reitmann and M. Schultz, Advanced Recurrent Neural Network based Aircraft Boarding Prediction. ATRS conference 2018
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 23
Prediction of passenger boarding progress using
neural network approach
Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR)
German Aerospace Center
Institute of Flight Guidance
Dr.-Ing. Michael Schultz
Head of Department Air Transportation
Phone +49 531 295-2570
Thank you.
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 24
Schultz, M. 2018. Implementation and Application of a Stochastic Aircraft Boarding Model. Journal of
Transportation Research Part C: Emerging Technologies 90, 334-349.
Schultz, M. 2017. Field Trial Measurements to Validate a Stochastic Aircraft Boarding Model. Aerospace
5(1), 27.
Schultz, M. 2017. Dynamic Change of Aircraft Seat Condition for Fast Boarding. Journal of
Transportation Research
Part C: Emerging Technologies 85, 131-147.
Schultz, M. 2018. A metric for the real-time evaluation of the aircraft boarding progress. Journal of
Transportation Research Part C: Emerging Technologies 86, 467-487.
S. Reitmann and M. Schultz, Advanced Recurrent Neural Network based Aircraft Boarding Prediction.
ATRS conference 2018
References
ICRAT • Castelldefels > Prediction of passenger boarding progress using neural network approach > June 2018 DLR.de • Chart 25