Logistique et Distribution

Facility Location

The Facility Location Problem

How to satisfy demands at a given number of
destinations with supplies from given set of distribution
centers.

Structure of the system is known

Location and demand of customers

Transportation means and costs

The characteristics of facilities

Goal: Select the most appropriate location of

distribution site(s) to satisfy demand from customers at
least cost.

1
Customer
DC

Customer
DC

2
Customer
DC

Customer
DC

3
Customer
DC

Customer
DC

4
 The Facility Location Problem
Various Applications
Location of an airport, a school, a plant, a

production facility, a post office, a hospital, a library

etc.
Single Facility or Multiple Facility Location Problem

Location decisions are important for customers :

They affect travel time to work, to administrative

offices, to supermarkets, to recreational centers, to

shopping malls.
Location decisions are important to business
organizations:
They affect the cost of doing business, and the flow of

goods and services.

Facility Location Models

Known or Unknown facilities
Number of Facilities (Single or Multiple)
Uncapacitated or Capacitated facilities
Continuous or Discrete Location Space
Distance Measures
Rectilinear, Euclidean, Actual distance

Objective: fixed cost plus transportation

Objective Functions
Minisum

Minimax

5
 Known Facility Location

EX1
Transportation Cost (Facilities known)
-- Transportation Problem

The ABC Ltd company has stores in 5 Midwestern cities. All

orders the product from the ABC Factory, which has
warehouses in 6 cities.
The demand for each of the 5 stores, the SKU available at
each of the 6 warehouse and the shipping costs (in dollars
per SKU) from warehouses to stores are also depicted bellow
Supply Capacity
Unit shipping cost ($/unit)
(SKU)
Minneapolis Salina Kansas Lincoln Wichita
Tulsa 3 9 6 5 7 77
Okla 6 4 5 4 5 45
Denver 7 3 7 1 2 85
St. Louis 3 7 9 2 2 50
Lawrence 4 3 3 8 4 94
Omaha 5 1 2 5 9 72
Demand (SKU) 52 99 83 71 65

6
 Mathematic Model
Indices and Sets
i, index for warehouses. i I = {Tulsa, Okla, Denver, St. Louis,
Lawrence, Omaha}
j, index for customers. j J = {Minneapolis, Salina, Kansas,
Lincoln, Wichita }
Parameters
cij: cost of transporting one unit from warehouse i to customer j
ai: supply capacity at warehouse i
bi: demand at customer j
Decision Variables
xij: number of units transported from warehouse i to customer j

Mathematic Model
Objective: Minimize Total Transportation Cost
m n
Z cij xij
i 1 j 1

Subject to
n

x
j 1
ij ai , i 1,2,..., m (supplyrestriction at warehouse i)
m

x
i 1
ij b j , j 1,2,..., n (demand requiremen t at market j)

xij 0, i, j 1,2,..., n (non - negativity restrictions)

7
 Transportation Problem with Lost Sales

Suppose, the warehouse in Omaha becomes unavailable.

Suppose the unit cost of unsatisfied demand is $30 for the
stores in Salina and Kansas, and $20 for the other
locations.

Transportation Problem with Lost Sales

Originally, the sum of supply was 423.

With Omaha gone, the total supply is now 351 units.

Since total demand is 370 units, 19 (=370-351) units of

demand will not be satisfied.
Replace Omaha by Lost Sales, with capacity equal to the
demand not satisfied, i.e., 19 units.

8
 Transshippement problem

Single Facility Location

9
Problem description

X Y Demand
A 2,5 4,5 2 000
B 2,5 2,5 5 000
C 5,5 4,5 10 000
D 5 2 7 000
E 8 5 10 000
F 7 2 20 000
G 9 2,5 14 000

Gravity Center Method

10
 Facility Location Problem (Continuous )

We need to find the location of a new facility (warehouse) to

serve several customers.

Suppose its location is (x, y). This is our decision variable

Suppose the location of customer i is Pi = (ai, bi) where i =

1m. Furthermore let wi be the weight associated with travel
between the new facility to the customers. This is our data

Objective: Minimize f (x) = sum (1m) wi d(X, Pi)

where d(X, Pi) could be rectilinear, Euclidean or Actual
distance.

Euclidean Distance
Objective: Minimize f (x) = sum (1m) wi d(X, Pi)
Where d (x, Pi) = sqrt( (x-xi)^2 + (y-yi)^2 )
F6=SQRT( (B6-$C$2)^2+(C6-$C$3)^2); copy an paste onto F7:F12
G6=D6*F6; copy an paste onto G7:G12
G13=SUM(G6:G12)

11
 Euclidean Distances

The total weighted sum of the distances is the sumproduct of the

forecasted demand and the Euclidean distances = 141,166.

Use Solver to Optimize the Location

12
 Here is the Solution !

Total weighted sum of the distances is reduced to

136,204.

Rectilinear Distance
Parallel to the x- and y- axes (east-west, north-south, and making 90 turns only.
F6=ABS(B6-$C$2)+ABS(C6-$C$3); copy an paste onto F7:F12
G6=D6*F6; copy an paste onto G7:G12
G13=SUM(G6:G12)

13
 Use Solver to Optimize the Location

Solver was able to reduce the total weighted sum of the distances based
on rectilinear distance from 180,147 to 161,000 or by about 10.6%!

14
100

80

60

40

20

0
0 20 40 60 80 100 120 140 160

An Example of Single Facilty Locaiton

City Demand
Reno is at 17, 55 100
Fort Worth is at 78, 20 90
Chicago is at 110, 65. 120

Find the Optimal SOluton to the above problem

Answer is (70.7, 48.7)
Using an Excel to solve the problem

15
100

80

60

40

20

0
0 20 40 60 80 100 120 140 160

100

80

60

40

20

0
0 20 40 60 80 100 120 140 160

16
 Minimax Location Problem
The objective is to minimize the maximum
distance
Why will you use a minimax locaiton model ??
If Rectilinear distance is used, this reduces to
min f ( x, y) max x ai y bi
1i m

If Euclidean distance is used, this reduces to

min f ( x, y ) max x ai y bi
2 2

1i m

MiniMax Euclidean Problem

Target Cell:
G13=MAX(G6-G12)

17
 MiniMax Rectilinear Problem
Target Cell:
G13=MAX(G6-G12)

Multiple Facility Location

18
 Simple Multi-Facility Location Problem

Consider the following problems

Two warehouses X1, X2 are set up to serve four facilities
P1 ( 5, 25)

P2 (25, 15)

P3 (10, 0)
P4 ( 0, 10)

It is also known that new ware X1 serves P1, P3, and P4

only, while X2 serves P2 and P3 only, In addition, materials
are transported between the two new warehouses.

Assume that cost are all proportion to travel distances, What

is the optimal locations of these facilities?

Continuous Multi-Facility Location Problem

Objective:
f (X1, X2) = v12d(X1,X2)
+ w11d(X1,P1) + w13d(X1,P3) +
w12d(X1,P4)
+ w22d(X2,P2) + w22d(X2,P3)
Let us assign v12 = w11 = w13 = w14 = w22 = w243)
Solve the problem using Euclidean and Rectilinear
distance

19
Multi-Facility Euclidean Distance

Multiple Facility Rectilinear

20
 Multi-Facility
Location and Allocation Model

The limitation of the above model

A) Weight can be hard to determine

B) If new facilities are of the same type, it does not

matter which NFs severs which EF.

Location-Allocation Problem
Where weights as well as the location of the facility are

decision variables.
Continuous Case, see paper
M.D. Devine and W.G. Lesso, Models for the Minimum

Cost Development of Offshore Oil Fields, Management

Science, Vol.18, 1972, pp. B378-287
L. Cooper, Location-Allocaiton Problems, Operations

Research, Vol 11. No.3, `963, pp. 331-334

Discrete Multi-Facility Location Problem

Covering Type Problem

PMedian Problem
PCenter Problem
Capacitated/uncapaciated Facility Locaiton
Problem

21
Customer
DC

Customer
DC

22
Where inventory needs to be for a 5 day order
response time - typical results --> 2 DCs

Customer
DC

Where inventory needs to be for a 3 day order

response time - typical results --> 5 DCs

Customer
DC

23
Where inventory needs to be for a next day order
response time - typical results --> 13 DCs

Customer
DC

Where inventory needs to be for a same day / next

day order response time - typical results --> 26 DCs

Customer
DC

24
 Location of a Citys Fire Stations

A city is made up of 11 of neighborhoods. A fire station can be placed in

any neighborhood and is able to handle the fires for both its neighborhood
and any adjacent neighborhood (any neighborhood with a non-zero border
with its home neighborhood).
The problem is to find the minimum number of fire stations used to cover
all neighborhoods.

Location of a Citys Fire Stations

Define Xi be the variable representing whether or not a first station
will be build in neighborhood i. If xi = 1, build; 0, otherwise

25
 Excel Implementation

Mathematic Models
Indices and Sets
i,j neighborhood indices, i = 111 Integer Program

Parameters
aij 1 neighboorhood I can be covered Min xi
by a fire station located at j i
0, otherwise s.t.
Decision Variables
xi = 1 if facility i is opened,
a x
j
ij j 1 for all i
0 otherwise
xi {0,1} for all i
The above problem can be formulated as

26
 Maximal Covering Problem
Now, suppose we can only set up p=2
fire stations, what is the maximal number Integer Program
of neighborhoods that can be covered
using p facilities. Max di yi
This is a p-covering problem. i
Model s.t.
xi = 1 if facility i is opened, 0 otherwise
yj = 1 if neighborhood j is covered, a x ij j yi for all i (*)
0 otherwise j
aij = 1 if neighborhood i can be covered
by a firestation at j. 0 otherwise
x
j
j p

Can you explain constraint (*) in English xi , y j {0,1} for all i, j

Solve the above problem using excel

p-Median/Center Problems

P-Center Problems
To locate n new facilities, called centers, on a network with

respect to m customers, so as to minimize the maximal distance

and a closet center
Example, located n emergence vehicles with in a neighborhood so

the maximum distance is minimized

P-Median Problem
To locate n new facilities, called medians, on a network with

respect to m existing facilities with vertex locations, so as to

minimize a sum of weighted distances between each existing
facility and a closet median.
Example: the set up of mail boxes where residence uses the

closest mailbox.

27
 p-Center Model
In p-center problem, we are minimizing maximum distance
between a market and a plant, or between fire stations and all
the houses served by the fire stations.

Replace the objective function in p-Median problem with

Min w
where w is
Max {dijxij : i is a market assigned to plant j}

P-center Model
Min w Minimize the Maximum Distance w

s.t.
y
jJ
j p There are a total of p centers

xij y j i I , j J If the center is setup, it can serve

xj
ij 1 i I Every one has to be served

w dij xij Definition of w: it is bigger than any

j

xij , y j {0,1} i I , j J Yes or no decision, integrity

28
 p-Median Model
Indices and Sets:
J: set of feasible plant locations, indexed by j
I: set of markets, indexed by i Min dij xij
Data i j
Di demand of market I
s.t.
No capacity limitations for plants
At most p plants are to be opened y
jG
j p
dij distance between market i and plant j
xij y j i I , j J
yj = 1 if plant is located at site j,
0 otherwise
xj
ij 1 i I

xij , y j {0,1} i I , j J
xij = 1 if market i is supplied from plant site j,
0 otherwise

Model Explanation
Min d ij xij Minimize the total weighted Distance w
i j

s.t. Constraints

y
jG
j p There are a total of p centers

xij y j i I , j J If the center is setup, it can sever

xj
ij 1 i I Every one has to be served

xij , y j {0,1} i I , j J Yes or no decision, integrity

Question:
How does the model guarantee a customer be served by a nearest center?

29
 Example Problem
Find the 2-median and 2-center solutions to the following 6 city problem
a) Write the mathematical model
b) Implement using excel solver

4
2 3
3 6 6

1 5 3
4
2 7
7
6 4

2 Median Problem Solution

30
 2 Center Problem

Optimal Solutions

Optimal solution to the 2 medina problem

Open 3 and 6

Total Distance: 13

Solution to the 2 center problem

Open 3 and 6

Maximal distance is 4

31
 The p-median & p-Center Problem
Assumptions
The number of warehouses to locate is fixed beforehand.

If a warehouse is located at site j:

There is no fixed cost,
There is no capacity limit.

Now, we relax this assumptions.

Uncapacitated Facility Location Problem

Suppose the cost of setting up the facility is different in different cities. In
the above 6 city problem, suppose the setup costs are (10, 8,15,11,8,12).
The transportation cost is 1 per unit distance.
How many warehouses should be setup, Where should we setup them,
and which customer do they serve?
Our goal is to minimize the total cost
4
2 3
3 6 6

5 3
1
4
2 6
7
6 4

32
 Uncapacited Facility Location Model

minimize f
jJ
j y j cij dij xij
iI jJ
Fixed cost +
Transportation Cost

subject to x
jJ
ij 1 i I Served at least once

xij y j i I , j J Open then serve

y j 0,1 j J
Yes/no decision
xij 0,1 i I , j J

Solution to the example problem

33
 A Facility Location Problem with Response Time

A new community is being developed and one of the decisions to be

made is where to allocate the two fire stations in the community.
For planning purposes, the community has been divided into five (5)
tracts, with no more than one fire station to be located in any given
tract. Each station is to respond to all of the fires that occur in the
tract in which it is located, as well as in the other tracts that are
assigned to the station. The following table gives the average
response time (in minutes) to a fire in each tract (columns) if their
tract is served by a station in a given tract (rows). The bottom row
gives the forecasted average number of fires that will occur in each
of tracts each day.
The objective is to minimize the overall average of the response
times to fires. How would you solve this problem?
Solve the problem using Excel Solver.

Response Time and Frequency Table

The table gives the average response time (in minutes) to a fire in each
tract (columns) if their tract is served by a station in a given tract (rows).

Station/Fire Track 1 Track 2 Track 3 Track 4 Track 5

Station 1 5 12 30 20 15
Station 2 20 4 15 10 25
Station 3 15 20 6 15 12
Station 4 25 15 25 4 10
Station 5 10 25 15 12 15
Frequency 2 1 3 1 3

The bottom row gives the forecasted average number of fires that
will occur in each of tracts each day.

34
 Mathematic Models
Indices
s, t: indices for tract T = {1,2,5}
Inputs
d(s,t): response time/ distance between s Min x( s, t ) d ( s, t ) f (t )
and t sT tT

f(t): frequency of fire in tract t. subject to

Decision Variables y s 2
x(s, t): 1: Fire in tract t will be covered sT
by station in tract s
0: Otherwise x ( s, t ) 1
sT
t T

y(s): 1, Build a fire station in track s x ( s, t ) ys s, t T

0, Otherwise
x( s, t ), y (t ) {0,1} s, t T
The problem can be can be formulated as

Model Explanation
The objective is to minimize the overall average
response time to fires and can be written as
s T, t T x (s, t) * d (s, t) * f (t)
The constraints are
There are a total of tow fire stations
Each tract has to be covered by a fire station so that
some one will respond to the fire.
If a tract doesnt allocate fire station then that tract
can not respond to the fire in any of the tracts
Binary: No more than one station can be allocated in
each tract,

35
 Computational Results
Stations Tracts
In track 1 2 3 4 5
1 (Yes) 1 1 0 0 0
2 (No ) 0 0 0 0 0
3 (yes) 0 0 1 1 1
4 (No ) 0 0 0 0 0
5 (No ) 0 0 0 0 0

Objective Value: Overall response time = 91 minutes

Solution: Fire stations are built in the tracts 1 and 3. Fire station
in tract 4 will respond to fires in tracts 1, 2 and fire station located
in tract 3 will respond to fires in tract 3, 4 and 5.

Excel Implementation
with a different optimal solution

36
 Problem
Now suppose that the following costs
(18,25,20,25,25) are incurred for building tracks
from 1 to 5.
Suppose that for each minutes is $ 1 dollar.
Relax the requirment that only 2 firestations are
built
Now, Solve the problem to find the minimum cost
to build the firestations and which track should
they serve.

Solve it using Excel

Capacitated Facility Location

The number of warehouses to locate is not fixed beforehand.

If a warehouse is located at site j:

A fixed cost fj is incurred,
There is a capacity qj on the amount of demand it
can serve.

We now have capacity constraints on the facility and therefore a retailer will
not always be assigned to its nearest warehouse.

37
 Capacitated Facility Location
i {customer}, j {facility}

min d e X f Y
i j
i ij ij
j
j j

subject to X 1
j
ij for all i

d X q Y i ij j j for all j
X ij 0,1
i
for all i, j
Y j 0,1 for all j

UFL
15 10 12 18
15 22 18 Numbers
A B C D
in red are
24 16 demands
12
20
18 E 5
25 24 F 24
12 12
G 15 H 30 I
11 16 13 22
19
22 25 19

J K L
19 21
22 19 20

38
 Multi-Source Capacitated Facility Locaiton

In the single-source CFLP, it is required that each retailer have only one
warehouse supplying it.
In the multiple-source CFLP, a retailers demand can be split between
any number of warehouses.

That is, a retailer may receive many deliveries of the same product from
different warehouses.

Other Problems
i {retailers}, j {warehouses}

min d e X f Y
i j
i ij ij
j
j j

subject to X 1
j
ij for all i

d X q Y
i
i ij j j for all j
X ij 0 for all i, j
Y j 0,1 for all j

39
 Multi-Level Location Allocation

We will now consider a model which takes into account the

transportation cost between plant and warehouse as well
as between warehouses and retailers.

A set of plants and retailers are geographically dispersed in

a region. Each retailers experiences demands for a product
that which are manufactured at the plants.

A set of warehouses must be located in the distribution

network from a list of potential sites.

Multi-Level Location & Allocaiton

i {retailers}, j {warehouses}, k {plants}

Inputs:

di = demand at retailer i
e1kj = cost of shipping one unit from plant k to
warehouse j
e2ji = cost of shipping one unit from warehouse j to
retailer i
fj = fixed cost of opening a warehouse at site j
sk = supply at plant k
qj = capacity of warehouse j

40
 Multi-Level Location & Allocaiton

Decision variables:

1 if a warehouse is located at site j

Yj
0 otherwise

1 if retailer i is served from warehouse j

X ji
0 otherwise

U kj amount of product shipped from plant k to warehouse j

Multi-Level Location & Allocaiton

min e1 U d e2
k j
kj kj
j i
i ji X ji f jY j
j
subject to X 1
j
ji for all i

d X q Y
i
i ji j j for all j

U d X
k
kj
i
i ji for all j

U s
j
kj k for all k

X ji 0,1 for all i, j

Y j 0,1 for all j
U kj 0 for all k , j

41
 Multi-Level Location & Allocaiton
min e1
k j l
kjl U kjl d il e2 jil X jil f jY j g k Pk
j i l j k

subject to X 1
j
jil for all i and l

v d X q Y
i l
l il jil j j for all j

U d X
k
kjl
i
il jil for all j and l

vU s P
j
l kjl k k for all k
X jil 0,1 for all j , i and l
Y j 0,1 for all j
Pk 0,1 for all k
U kjl 0 for all k , j and l

Discrete Multi-Facility Location Problems

Location Covering Problems

To find the minimum number of facilities to

cover all customers and the distance between

facilities and customer should not exceed a
specified limit.
P-Median Problem
To locate n new facilities, called medians, on a

network with respect to m existing facilities

with vertex locations, so as to minimize a sum
of weighted distances between each existing
facility and a closet median.
P-Center Problems
To locate n new facilities, called centers, on a

42
 Discrete Multi-Facility Location Problems

UFL or CFL -- Warehouse location problem

In this problem, we have m customers, each

with a known delivery demand. The are p

potential warehouse locations, with an fixed
cost of establishing a warehouse.
The problem is to determine which warehouse

sites should be used and the assignment of

customers to warehouses, in order to minimize
the total cost.

Thanks

43