Applications of Linear
Programming
Dr. Antonio A. Trani
Associate Professor of Civil Engineering
Virginia Tech
Analysis of Air Transportation Systems
Falls Church, Virginia
NEXTOR/Virginia Tech - Air Transportation Systems Lab 1
Recall the Standard LP Form
n
Maximize ∑ cj xj
j=1
n
subject to: ∑ aij xj ≤ bi for i = 1, 2, …, m
j=1
x j ≥ 0 for j = 1, 2, …, n
Suppose that now we wish to investigate how to handle
Minimization problems:
NEXTOR/Virginia Tech - Air Transportation Systems Lab 2
Minimization Problem
Formulation
n
Minimize z=
∑c x j j
j=1
Reformulate as:
n
Maximize –z = –
∑c x j j
j=1
NEXTOR/Virginia Tech - Air Transportation Systems Lab 3
Airline Scheduling Problem (ASP-1)
A small airline would like to use mathematical
programming to schedule its flights to maximize profit.
The following map shows the city pairs to be operated.
1 3
New York
Cincinnati
λ23 = 450 pax/day
λ32 = 500 pax/day
λ21 = 600 pax/day d23 = 375 nm
λ12 = 450 pax/day
2 Roanoke
d12 = 260 nm λ24 = 760 pax/day
λ42 = 700 pax/day
4 d24 = 310 nm
λij = Demand from Atlanta
i to j
NEXTOR/Virginia Tech - Air Transportation Systems Lab 4
Airline Scheduling Problem
The airline has decided to purchase two types of aircraft
to satisfy its needs: 1) the Embraer 145, a 45-seat
regional jet, and 2) the Avro RJ-100, a four-engine 100
seater aircraft (see the following figure).
EMB-145
Avro RJ-100
NEXTOR/Virginia Tech - Air Transportation Systems Lab 5
Aircraft Characteristics
Aircraft EMB-145 Avro RJ-100
Seating capacity - n k 50 100
Block speed (knots) - v k 400 425
Operating cost ($/hr) - ck 1,850 3,800
Maximum aircraft utili- 13.0 12.0
zation (hr/day)a - Uk
a. The aircraft utilization represents the maximum number of hours an aircraft is in actual use with the engines
running (in airline parlance this is the sum of all daily block times). Turnaround times at the airport are not
part of the utilization variable as defined here.
NEXTOR/Virginia Tech - Air Transportation Systems Lab 6
Nomenclature
Define the following sets of decision variables:
No. of aircraft of type k in fleet = Ak
No. flights assigned from i to j using aircraft of type k = Nijk
Minimum flight frequency between i and j = (Nij)min
NEXTOR/Virginia Tech - Air Transportation Systems Lab 7
Based on expected load factors, the tentative fares
between origin and destination pairs are indicated in the
following table.
Average one-
Origin-
City pair designator way fare
Destination ($/seat)
ROA-CVG Roanoke to 175.00
Cincinnati
ROA-LGA Roanoke to La 230.00
Guardia
ROA-ATL Roanoke to 200.00
Atlanta
NEXTOR/Virginia Tech - Air Transportation Systems Lab 8
Problem # 1 ASP-1 Formulation
1) Write a mathematical programming formulation to
solve the ASP-1 Problem with the following constraints:
Maximize Profit
subject to:
• aircraft availability constraint
• demand fulfillment constraint
• minimum frequency constraint
NEXTOR/Virginia Tech - Air Transportation Systems Lab 9
Problem # 2 ASP-1 Solution
1) Solve problem ASP-1 under the following numerical
assumptions:
a) Maximize profit solving for the fleet size and frequency
assignment without a minimum frequency constraint. Find
the number of aircraft of each type and the number of flights
between each origin-destination pair to satisfy the two basic
constraints (demand and supply constraints).
b) Repeat part (a) if the minimum number of flights in the
arc ROA-ATL is 8 per day (8 more from ATL-ROA) to
establish a shuttle system between these city pairs.
NEXTOR/Virginia Tech - Air Transportation Systems Lab 10
c) Suppose the demand function λ ij varies according
to the number of flights scheduled between city pairs (see
the following illustration). Reformulate the problem and
explain (do not solve) the best way to reach an optimal
solution.
λij (λij) max
(λij) min
Nij
NEXTOR/Virginia Tech - Air Transportation Systems Lab 11
Vehicle Scheduling Problem
Formulation of the problem.
Maximize Profit
subject to: (possible types of constraints)
a) aircraft availability constraint
b) demand fulfillment constraint
c) Minimum frequency constraint
d) Landing restriction constraint
NEXTOR/Virginia Tech - Air Transportation Systems Lab 12
Vehicle Scheduling Problem
Profit Function
P = Revenue - Cost
Revenue Function
Revenue = ∑λ f ij ij
(i j)
where: λ ij is the demand from i to j (daily demand)
f ij is the average fare flying from i to j
NEXTOR/Virginia Tech - Air Transportation Systems Lab 13
Vehicle Scheduling Problem
Cost function
let N be the flight frequency from i to j using aircraft
ijk
type k
let C be the total cost per flight from i to j using
ijk
aircraft k
Cost = ∑∑N ijk C ijk
(i j) k
then the profit function becomes,
NEXTOR/Virginia Tech - Air Transportation Systems Lab 14
Profit = ∑ λ f –∑ ∑ N
i, j
ij ij
i, j k
ijk C ijk
NEXTOR/Virginia Tech - Air Transportation Systems Lab 15
Vehicle Scheduling Problem
Demand fulfillment constraint
Supply of seats offered > Demand for service
∑n N k ijk ≥ λ ij for all ( i, j ) city pairs or alternatively
k
∑ ( lf )n N k ijk ≥ λ ij for all ( i, j ) city pairs
k
lf is the load factor desired in the operation (0.8-0.85)
Note: airlines actually overbook flights so they usually
factor a target load factor in their schedules to account
for some slack
NEXTOR/Virginia Tech - Air Transportation Systems Lab 16
Vehicle Scheduling Problem
Aircraft availability constraint
(block time) (no. of flights) < (utilization)(no. of
aircraft)
∑t ijk N ijk ≤ U k A k
(i j)
one constraint equation for every k aircraft type
NEXTOR/Virginia Tech - Air Transportation Systems Lab 17
Vehicle Scheduling Problem
Minimum frequency constraint
No. of flights between i and j > Minimum number of
desired flights
∑ Nijk ≥ ( Nij )min for all ( i, j ) city pairs
k
Note: Airlines use this strategy to gain market share in
highly traveled markets
NEXTOR/Virginia Tech - Air Transportation Systems Lab 18
Vehicle Scheduling Problem
Maximize Profit = ∑ λ f –∑ ∑ N
i, j
ij ij
i, j k
ijk C ijk
subject to
∑n N k ijk ≥ λ ij for all ( i, j ) city pairs
k
∑t ijk N ijk ≤ U k A k for every k aircraft type
(i j)
∑ Nijk ≥ ( Nij )min for all ( i, j ) city pairs
k
NEXTOR/Virginia Tech - Air Transportation Systems Lab 19
Demo with Excel
NEXTOR/Virginia Tech - Air Transportation Systems Lab 20
Crew Scheduling Problem
A small airline uses LP to allocate crew resources to
minimize cost. The following map shows the city pairs
to be operated.
1 New York
Denver 3
Morning
2 Dallas Afternoon
Night
DEN = Denver, 4 MEX = Mexico City,
DFW = Dallas Mexico JFK = New York
NEXTOR/Virginia Tech - Air Transportation Systems Lab 21
Crew Scheduling Problem
Flight Number O-D Pair Time of Day
100 DEN-DFW Morning
200 DFW-DEN Afternoon
300 DFW-MEX Afternoon
400 MEX-DFW Night
500 DFW-JFK Morning
600 JFK-DFW Night
700 DEN-JFK Afternoon
800 JFK-DEN Afternoon
NEXTOR/Virginia Tech - Air Transportation Systems Lab 22
Crew Scheduling Problem
Definition of terms:
a) Rotations consists of 1 to 2 flights (to make the
problem simple)
b) Rotations cost $2,500 if terminates in the originating
city
c) Rotations cost $3,500 if terminating elsewhere
Example of a feasible rotations are (100, 200),
(500,800),(500), etc.
NEXTOR/Virginia Tech - Air Transportation Systems Lab 23
Crew Scheduling Problem
Single
Ri Ri Two-flight
Flight Cost ($) Cost ($)
Rotations
Rotations
1 100 3,500 9 100,200 2,500
2 200 3,500 10 100,300 3,500
3 300 3,500 11 500,800 3,500
4 400 3,500 12 500,600 2,500
5 500 3,500 13 300,400 2,500
6 600 3,500 14 200,100 3,500
7 700 3,500 15 600,300 3,500
NEXTOR/Virginia Tech - Air Transportation Systems Lab 24
Single
Ri Ri Two-flight
Flight Cost ($) Cost ($)
Rotations
Rotations
8 800 3,500 16 600,200 3,500
17 600,500 3,500
18 800,100 3,500
19 700,600 3,500
20 700,800 3,500
Decision variables:
⎛ 1 if i rotation is used
Ri = ⎜
⎝ 0 if i rotation is not used
NEXTOR/Virginia Tech - Air Transportation Systems Lab 25
Crew Scheduling Problem
Min Cost
subject to: (possible types of constraints)
a) each flight belongs to a rotation (to a crew)
Min
z = 3500 R1 + 3500 R2 + 3500 R3 + 3500 R4 + 3500
R +
5
3500 R6 + 3500 R7 + 3500 R8 + 2500 R9 + 3500 R 10
NEXTOR/Virginia Tech - Air Transportation Systems Lab 26
3500 R 11 + 2500 R 12 + 2500 R 13 + 3500 R 14 +
3500 R 15
3500 R 16 + 3500 R 17 + 3500 R 18 + 3500 R 19 + 3500
R 20
s.t. (Flt. 100) R1 + R9 + R 10 + R 14 + R 18 =1
(Flt. 200) R2 + R9 + R 14 + R 16 =1
(Flt. 300) R3 + R 10 + R 13 + R 15 =1
(Flt. 400) R4 + R 13 =1
(Flt. 500) R5 + R 11 + R 12 + R 17 =1
NEXTOR/Virginia Tech - Air Transportation Systems Lab 27
(Flt. 600) R6 + R 12 + R 15 + R 16 + R 17 +
R =1
19
(Flt. 700) R7 + R 19 + R 20 =1
(Flt. 800) R8 + R 11 + R 18 + R 20 =1
NEXTOR/Virginia Tech - Air Transportation Systems Lab 28
Crew Scheduling Problem
Problem statistics:
a) 20 decision variables (rotations)
b) 8 functional constraints (one for each
flight)
c) All constraints have equality signs
NEXTOR/Virginia Tech - Air Transportation Systems Lab 29
Crew Scheduling Problem
(Solver)
Demo in Excel
NEXTOR/Virginia Tech - Air Transportation Systems Lab 30
Human Resource Assignment
Problem (ATC Application)
Linear programming problems are quite useful for solving
staffing problems where human resources are typically
scheduled over periods of varying activities. Consider the
case of the staffing requirements of a busy Air Route Traffic
Control Center (ARTCC) where Air Traffic Control (ATC)
personnel monitor and direct flights over large regions of
airspace in the Continental U.S.
Given that traffic demands vary over the time of day ATC
controller staffing requirements vary as well. Take for
example Jacksonville ARTCC comprised of 35 sector
NEXTOR/Virginia Tech - Air Transportation Systems Lab 31
boundaries (see the Figure below). Each sector is
managed by one or more controllers depending on the
traffic load.
NEXTOR/Virginia Tech - Air Transportation Systems Lab 32
ATC Resource Allocation
Jacksonville ARTCC Sectorization at 40,000 ft.
Sectors at FL400
36
35 Washington
Atlanta Enroute Center
34 113
Enroute Center 91
131
Latitude (deg.)
33
126
32 129
93 51
31
61
87 Jacksonville
86
115
121
30 62
Enroute
89
Center49
119
Sector
23 77 143
40
24 55
29
84 75
28
Miami
27 Enroute Center
26
-88 -86 -84 -82 -80 -78 -76
Longitude (deg.)
NEXTOR/Virginia Tech - Air Transportation Systems Lab 33
Relevant Questions
A task analysis study estimates the staffing requirements for
this ARTCC (see Table 1). Let x be the number of ATC
i
controllers that start their workday during the i th hour
( x , …, x).
2
a) Formulate this problem as a linear programming
problem to find the least number of controllers to satisfy
the staffing constraints based on traffic demands expected at
this FAA facility. Assume controllers work shifts of 8 hours
(no overtime is allowed for now).
NEXTOR/Virginia Tech - Air Transportation Systems Lab 34
b) Find the minimum number of controllers needed
to satisfy the staffing requirements using Excel or Matlab.
Comment on the solution obtained.
c) Human factors studies suggest ATC controllers take one
hour of rest during their 8-hour work period to avoid
excessive stress. The ATC manager at this facility instructs
all personnel to take the one-hour rest period after working
four consecutive hours. Reformulate the problem and find
the new optimal solution.
d) The average salary for ATC personnel is $100,000 for
normal operation hours (5:00 -19:00 hours) with a 15%
higher compensation for those working the night shift
(19:00 until 5:00 hours). Reformulate the problem to
NEXTOR/Virginia Tech - Air Transportation Systems Lab 35
allocate ATC controllers to minimize the cost of the
operation. Assume the one-hour break rule applies.
TABLE 1. Expected Staffing Requirements at Jacksonville ARTCC Center
(Jacksonville, FL).
Time of Day (EST) Staff Needs Remarks
0:00 - 2:00 30 Light traffic
2:00 - 5:00 25 Light traffic - few air-
line flights
5:00 - 7:00 35 Moderate traffic
7:00 - 10:00 48 Heavy traffic (morning
“push”)
10:00 - 13:00 35 Moderate traffic
13:00 - 17:00 31 Moderate traffic
17:00 - 21:00 42 Heavy evening traffic
21:00 - 24:00 34 Moderate traffic
NEXTOR/Virginia Tech - Air Transportation Systems Lab 36
