Agent-Based Computational
Approach to Airline
Competition and Airport
Congestion Problems
Junhyuk Kim
Dušan Teodorovic
Antonio Trani
Virginia Tech
Agent-Based Approach
n
The traditional approach to analyze
transportation problems has been the “Topdown approach”
n
The individual parts (passengers, airlines,
airports, and Air Transportation authorities)
have great autonomy to make decisions,
communicate and to interact with one another
2
Agents
3
Objectives of Agents
n
Airline : Maximize Profit
n
Passenger : Decrease Travel time and Cost
n
Airport : Minimize Congestion
4
Agent Behavior
n
Relatively minor change in agents’ behavior
can cause significant change in whole system
n
Complex collective behavior can be emerged
from simple actions of individual agents passengers, airlines, airports and aviation
authorities
5
Simulate Agent Behavior
n
Congestion in air transportation could be
viewed as emergent phenomena that is
sometimes difficult to predict and that is even
sometimes counterintuitive
n
A promising way for analyzing congestion in
air transportation is the development of the
simulation models that can simulate behavior
of every agent
6
Agent-Based Model in Air Transportation
n
Our Agent-based model considers that each
part act based on its local knowledge and
competes and/or cooperates with other parts
n
Model developed allows agents that represent
airports to increase the capacity, or to
significantly change landing fee policy, while
the agents that represent airlines learn all the
time, change their markets, fares structure,
flight frequencies, and schedules
7
Congestion Pricing
The basic idea of congestion pricing in air
transportation is to introduce peak-period
pricing and other possible strategies as a
powerful tool that will be capable to
modify airlines' use of existing airways and
airports and to change passengers' behavior
8
Congestion Pricing (Continued)
n
Vickrey: “Charges should reflect as closely as
possible the marginal social cost of each trip
in terms of the impacts on others. There is no
excuse for charges below marginal social
cost.”
n
Airlines (and passengers) should pay a price
equivalent to the delay cost they impose on
others
9
Non-Cooperative Evolutionary Game
between Agents
To explore evolved unplanned coordination
under the different landing pricing strategies
produce similar results like the planned global
coordination with the “central planner”
whose main objective is the minimization of
the total air traffic congestion
10
Airlines Behavior
For every airline that operates in the network
determine:
n
Set of routes that airline flies, aircraft types on
these routes, flight frequencies and departure
times
n
Under the time dependent airport landing fees
11
Airlines Competition
8
9
2
10
7
4
1
5
3
Airlines A
Airlines B
6
12
Airlines C
Strategy of Airlines
The strategy of airline j related to the i-th market
during G-th iteration
:
SijG = {PijG , N ijG , AijG , DijG }, i = 1, 2,...,K
13
j = 1, 2, ..., M
Payoff of Airlines
The total payoff of airline j at generation G :
TP
G
j
=
M
∑
i
14
P (S
G
ij
)
Payoff of Airlines (Continued)
n
n
n
n
15
Iteration (generation) represents certain time unit
Airlines collect information about the profit they
make, as well as information about activities of
their competitors
Based on this information airlines change
markets, aircraft types, flight frequencies, and
departure times
Airlines change through the evolution, from
iteration to iteration, their operating strategy
Representation of Airline Strategies
Time Slot
Time Slot
Route 1 :
0
1
0
0
3
0
0
………….
0
2
Route 2 :
0
0
0
0
0
0
0
………….
0
0
Route 3 :
0
0
0
1
3
0
0
………….
0
2
Route 4 :
0
3
0
0
1
0
0
………….
0
0
0
3
0
0
1
0
0
0
0
0
0
0
0
………….0
Route 2 :
0
1
0
0
0
0
0
………….0
Route 3 :
0
0
0
0
0
0
0
………….0
Route 4 :
0
3
0
0
0
0
0
………….0
Route i :
0
0
0
2
0
0
0
………….0
………..
0
Airline 1
16
………..
Route1 :
………
………
Route i :
………..
………….
0
0
………..
Airline j
What we can determine
n
Markets that airline serve
n
Flight frequencies on all airline’s routes
n
Fleet assignment
n
Aircraft departure times on all routes
17
Flight (Travel option) choice:
Pass/Hr
i
18
k
Time
Passenger’s Flight Choice based
on Logit model
p
ik
=
U
e
U
∑ e
f ∈ F
19
ik
if
Passenger’s Flight Choice based
on Logit model (Continued)
Utility associated with the passenger from the i-th time slot
who chooses flight that departs in the k-th time slot
U ik = a ⋅ Tik + b ⋅ ATk
20
:
Fuzzy Logic for Flight Choice Model
n
If
Then
21
schedule delay is LOW and ticket
price is ACCEPTABLE and total
travel time is SHORT
passenger’s preference to choose
the considered flight is VERY
HIGH
Evolutionary Strategies
n
Various evolutionary strategies (mimic
strategies) are applied to compare profits of
all airlines and also convergence ability
n
Most of evolutionary strategies that we have
explored are based on the logic that airlines in
some way change flight schedule on the low
performance markets
22
Model Procedure
(1-1)
O-D DEMAND
MATRIX
START
(2-1)
FLIGHT
SCHEDULING
(2-2)
PRICE
SCHEDULING
(1-2)
PARTICIPITATION
MATRIX
(1) INPUT DATA
(1-3)
MARKET-ROUTE
MATRIX
(2) DEVELOP INITIAL
STRATEGY FOR EACH
AIRLINE
(1-4)
ROUTE-LEG
MATRIX
(1-5)
LANDING FEE
MATRIX
(TIME OR WEIGHT)
(3) EVOLVE
AIRLINE & AIRPORT
STRATEGY
PASSENGER STRATEGY
(1-5)
AIRPORT
CAPACITY MATRIX
(4) SIMULATE
PASSENGERS' BEHAVIOR
TO CHOOSE FLIGHT
(4-1)
LOGIT MODEL
FOR FLIGHT
CHOICE
(4-2)
FUZZY LOGIC
(5) SIMULATE
FLIGHTS IN
PRE-DEFINED NETWORK
AIRPORT STRATEGY
AIRLINE STRATEGY
(6) COMPUTE PROFIT
OF EACH AIRLINES
(3-4)
LANDING FEE
BY
LANDING TIME
(3-1)
STOCHASTIC
MIMIC
(3-2)
DETERMINISTIC
MIMIC
(3-5)
LANDING FEE
BY
AIRCRAFT
WEIGHT
(7) SAVE SIMULATION
RESULTS
(3-3)
PERTURBATE
FLIGHT
SCHEDULE
(3-6)
SLOT AUCTION
(8) FINAL ITERATION?
YES
TERMINATE
23
No
Sample Problem
1
6
7
2
3
8
HUB I
HUB II
4
9
5
24
10
c arrier1
c arrier2
c arrier3
Deterministic mimic strategy
195000
190000
185000
Average Profit
($) 180000
175000
170000
165000
160000
155000
150000
145000
140000
1
3
5
7
9
11
13
15
17
Generation
25
19
21
23
25
27
29
Disturbing flight schedule strategy
195000
carrier1
carrier2
carrier3
190000
185000
Average Profit
($) 180000
175000
170000
165000
160000
155000
150000
145000
140000
1
13 25 37 49 61 73
85 97 109 121 133 145 157 169 181 193
Generation
26
Open questions
n
Would different landing pricing strategies change
the runway occupancy rate or average parking
duration?
n
Would airlines easily accept new landing pricing
strategies?
n
What is the most appropriate fare structure when
introducing different landing pricing strategies?
27
Open questions (Continued)
n
Would proposed systems significantly decrease the
total number of flights during peak hours and
increase the total number of flights outside peakhours?
n
How will regional carriers behave vs large
carriers?
n
How will certain classes of air passengers behave
in the situation when direct operating cost and ticket
prices increase for the peak-hour flights?
28