18.4.2017.

TRANSPORTATION PROBLEM

Overview
• The transport problem or model of transport first
was set by F. L. Hitchcock, 1941
• Before him similar problems were examined by L.V.
Kantorovich
• The problem of transport was set and resolved
before the linear programming was descovered, and
later it turned out that the problem of transport is
actually a special case of LP and can be solved using
the simplex method

1
 18.4.2017.

Overview
• We are considering a situation where we have a certain number of
warehouses (origins) with goods from which we need to supply a
certain number of consumer centers (destinations)
• From each warehouse we can ship goods to every consumer center
• We set the requirement that the total transportation cost should be
minimal
• Beside this, transport model can have other applications
• Every problem that meets the requirements of transport model can
be solved using transport methods,
• The problem of transport is usualy a problem of minimum, but it
can be formulated as a problem of maximum (to maximize profit or
revenue)

Assumptions of the model

1. We have m warehouses or the origins from which the goods are
transported to n consumer centers or destinations
2. Goods are homogeneous, and that means that for the
destination is the same from which origin the goods are coming
3. The unit cost of transport or transport rates are constant and do
not depend on the total quantity of goods that are shipped.
4. The total amount of goods at the origins is equal to the total
demand of goods in destinations
• Often the third assumption limits the applicability of the model in
practice
• If the fourth assumption is not satisfied, this can be corrected
easily (to transform the „open" model into the „closed” model)

2
 18.4.2017.

Setting up the problem

Setting up the problem

• Oi - Destination i
• Ij - Origin j
• ai - Available quantities of goods at the origin i
• bj - The required quantity of goods at the destination j
• cij - Unit price (cost of transportation) from the origin i to
the destination j (If this is a problem of maximum, then
this can be the revenue or profit that can be earned from
the shippment of one unit from the origin i to the
destination j)
• Xij - Total goods shipped from the origin i to the
destination j

3
 18.4.2017.

Setting up the problem

• The model can be opened or closed. If it is open it is
necessary to close it before solving it
• If Σai = Σbj then the problem is closed
• If Σai < Σbj then the problem is open. We close it by
introducing artifical (dummy) origin which will
contain Σbj – Σai of goods
• If Σai > Σbj then the problem is also open. We close it
by introducing artifical (dummy) destination that
will requiered Σai – Σbj of goods.

Setting up the problem

• The cost of transport cj from artificial origins to all
destinations, and from all origins into artificial
destination are 0 because this transport actually will
not happen
• To solve the model of transport it means to
determine the transportation plan – what amount of
goods should be shipped from every single origin to
every single destination.
• It means to determine the value of variables xij that
will make the transport cost (Z) to be minimized (or
maximized)

4
 18.4.2017.

Transportation problem as LP problem

• Based on Table from slide 5 we can setup the transportation problem like LP model
• The function of objective will be the total cost of transportation, and the constraints are
based on the requirement that from each origin must be transported total quantity of goods
and that each destination has to be completely satisfied:

m n
min Z   i 1 j 1
c ij  x ij

x 11  x 12    x 1 n  a1
x 21  x 22    x 2 n  a2
...
x m 1  x m 2    x mn  a m
x 11  x 21  x m1  b1
x 12  x 22  xm2  b2
...
x1 n  x2n  x mn  b n
x ij  0

10

Transportation problem as LP problem

• We have LP model with m × n variables and m + n constraints.
• All restrictions are with the sign "=", which means that it is necessary
to use M procedure
• We can define a dual model that is very important because MODI
method for optimization of the initial solution is based on it
• Dual will be a problem of maximum and it will have m × n constraints
and m + n variables

m n
maxv    ri   s j
i 1 j 1

ri  s j  cij , i  1, 2,  , m; j  1, 2,, n

5
 18.4.2017.

11

Example – Foster Generators

• Foster Generators operates plants in Cleveland,
Ohio; Bedford, Indiana; and York, Pennsylvania.
Production capacities over the next three-month
planning period for one particular type of generator
are as follows:

12

• The firm distributes its generators through four

regional distribution centers located in Boston,
Chicago, St. Louis, and Lexington; the three-month
forecast of demand for the distribution centers is as
follows:

6
 18.4.2017.

13

• Management would like to determine how much of its

production should be shipped from each plant to each
distribution center that will provide the minimum total
transportation cost.
• The cost for each unit shipped on each route is given in
Table

14

The network representation of the foster generators transportation

problem

7
 18.4.2017.

15

Transportation table for the foster generators transportation problem

16

Transportation methods

• The transport problem can be solved using two groups of

methods:
• Methods for determining the inital possible solution
▫ Method North - West Corner or Upper - Left Corner
▫ The Method of Preferred Flows or Minimum Cost Method
▫ Vogel's Aproximation Method
• Methods for optimizing the solution
▫ Stepping-stone Method or Method of Jumping from Stone
to Stone
▫ MODI Method or a Modified Distribution Method

8
 18.4.2017.

17

Methods for determining the inital possible

solution
• Method North-west Corner or Upper-left Corner
• Method of Preferred Flows or Minimum Cost
Method
• Vogel's Aproximation Method

18

The North-west corner method or upper-left corner

• This method does not take into account the cost of transport (cij)
• It suggests to allocate as much flow as possible to a cell located upper-left (or
north-west) regardless of cij
• This method is the simpliest one, but often leads us far away from the optimal
solution, and therefore it is not recommended to be used.
• So, in this method the allocation of units starts from the northwest corner, i.e.
from the field (1.1), in which we can place as many units as we have in the first
origin, or as many it is reqiered of the first destination.
• If b1<a1 then x11 = b1
• If b1>a1 then x11 = a1
• In this way the supply of the first origin is exhausted, or the demand of the
first destination is settled, and this origin (or this destination) should be
excluded from further allocation.
• The process continues with the remaining cells until we allocate all resources
to satisfy all requirements of the destinations

9
 18.4.2017.

19

Example - Initial solution obtained using the North-west corner

method for the Foster Generators transportation problem

20

The Method of Preferred Flows or Minimum

Cost Method
• Initial solution is obtained by allocating as much as possible to
the most favorable cell in matrix, and that is cell with the
smallest cij (if we are minimizing Z), i.e. the highest cij (if we
are maximizing Z).
• With this, the supply of some origin will be exhausted, or the
demand of some destination will be satisfied. That row
(origin) or column (destination) is „excluded" from the further
allocation.
• The procedure should be repeated until all available units are
assigned to certain destinations.
• Because this method takes into account the cost of transport
cij, this method generally leads us closer to the optimal
solution than the upper left corner method.

10
 18.4.2017.

21

Example - Initial solution obtained using the Minimum cost

method for the Foster Generators transportation problem

22

Vogel’s aproximation method

• This method is more complex for the usage than previous two methods,
but usualy brings us closest to the optimal solution.
• That is why it is recommended to be used
• Process of finding the initial feasible solution for the problem of minimum
and the problem of maximum is different.
• If we minimize:
• 1) Calculate the difference between two lowest cij for the each row and
each column.
• 2) Choose the row or column with a maximum difference
• 3) In the most favorable cell (cell with the smallest cij ) of the chosen row
or column, the maximum possible number of units (xij) should be
allocated.
• With this, the supply of some origin will be exhausted, or the demand of
some destination will be satisfied. That row (origin) or column
(destination) is „excluded" from the further allocation.

11
 18.4.2017.

23

Vogel’s aproximation method

• The procedure shoud be repeated until all available units are
assigned to certain destinations.
• When the allocation is reduced to a single row or a single
column, the differences can not be calculated, therefore
allocation is performed on the fields of the remaining column,
or row.
• If we maximize:
• 1) Calculate the difference between two largest cij for the
each row and each column.
• 2) Choose the row or column with a maximum difference
• 3) In the most favorable cell (cell with the largest cij ) of the
chosen row or column the maximum possible number of units
(xij) should be allocated.

24

Example - Initial solution obtained using the Vogel’s aproximation

method for the Foster Generators transportation problem

12
 18.4.2017.

25

Methods for optimizing the initial feasible

solution
• The optimal solution is achieved by iterative
improvement of the initial feasible solution.
• Methods for the optimization are:
• Stepping Stone Method or Methods of Jumping from
Stone to Stone
• MODI method or Modified Distribution Method

26

Stepping-stone Method or Method of

jumping from stone to stone
• The initial solution is tested by calculation of the relative costs
(differences) dij of unoccupied cells in the matrix of transport.
• To calculate dij we conduct the alternating addition and subtraction of
unit costs (cij), starting from the cell for which we calculate relative
cost, continuing further over the cells located at the path of
alternating „jumping” from a „stone” in a row, to a „stone” in a
column, until we return back to the starting point cell.
▫ Note: Stone is xij, ie. a certain number of units at an occupied cell.
• dij with the sign "-" shows that the objective function can be improved
if we minimize
• dij with the sign "+" shows that the objective function can be improved
if we maximize.

13
 18.4.2017.

27

Stepping-stone Method or Method of

jumping from stone to stone
• How to change the initial solution:
• 1) Select the cell with the largest negative dij if you minimize, or
with the largest positive dij if you maximize
• 2) At the cell from point 1) the minimum value, which is located in
some occupied cell whose cij was substracted in a process of
calculation of dij, should be allocated
• 3) Other cells should remain the same or they should be corrected
by adding or subtracting the amount determined in step 2)
▫ Note: After the achieving of new solution it is recomended to check
whether the sum of the values xij in rows and columns still matches the
supply of sources and demand of destinations.
• The new value of objective function Z will decrease in the case of
minimization, or increase in case of the maximization

28

Stepping-stone Method or Method of

jumping from stone to stone
• We obtain a new solution that is better than the previous
one, but we do not know if there exists even better solution.
• It is necessary to examine for a new solution whether it is the
optimal solution, or not, by recalculating dij.
• The optimal solution is achieved
• For a minimum: when all the dij are positive or 0
• For a maximum: when all the dij are negative or 0
• The existence of dij values equal to 0 in the final solution
means that there are alternative optimal solutions, i.e. we can
occupy these cells and the objective function will not change
(remains optimal).

14
 18.4.2017.

29

Example - Optimization using Stepping-stone Method for

the Foster Generators transportation problem

• We start with initial solution obtained with Vogel’s aproximation

method
• We have to check wether we should occupy cells (1,3), (1,4),
(2,2), (3,2), (3,3) and (3,4) by calculating the difference for these
cells:
• d1,3 = 7-3+7-2= 9
• d1,4 =6-3+7-3= 7
• d2,2 =5-7+3-2= -1 – negative sign means that we should occupy cell (2,2)
• d3,2 =5-2+3-2= 4
• d3,3 =4-2+7-2= 7
• d3,4 =5-2+7-3= 7

30

• To asign some quantity to cell (2,2) we have to reduce the quantities on cells (1,2)
and (2,1)
• The maximum amount that we can take from these cells is min(2500, 4000) and
that is 2500
• New solution will look like this

15
 18.4.2017.

31

• Now we have to try to improve this solution. Again, we are going to use stepping
stone method;
• d1,3 = 7-2+5-2= 8, d1,4 =6-2+5-3= 6, d1,2 =7-5+2-3= 1
• d3,2 =5-2+3-2= 4, d3,3 =4-2+3-2+5-2= 6, d3,4 =5-2+3-2+5-3= 6
• Obtained values for dij (all positive - we try to minimize our goal function)
indicate that this is final and optimal solution (without alternative solutions)

32

• The final solution indicates following:

• We shoud ship from Cleveland 3500 units to Boston,
and 1500 to Chicago.
• Furthermore, we shold ship from Bedford 2500 units
to Chicago, 2000 to St. Louis and 1500 to Lexington
• Finally, the solution suggests that we should ship
2500 units to Boston
• The total cost of shippment will be 39500
• Following Table summarizes the optimal solution and
shows total cost

16
 18.4.2017.

33

Optimal solution to the Foster Generators

transportation problem

34

MODI Method or a Modified Distribution

Method
• This method is based on the dual model
• The initial solution is also tested by calculation of difference (dij), but this
difference is obtained in a different way (compared to the stepping stone
method):
• 1) Using the formula ri+sj=cij we set an equation for all the occupied cells in
order to construct a system of m+n–1 linear equations with m+n unknowns.
• ri and sj are unknowns.
• There are ri unknowns as many as there are origins, and there are sj
unknowns as many as there are destinations
• In the obtained system equations there is at least one equation less than we
have unknowns, and for the problem with the degeneration we could have
two (or even more).
• In order to solve the system of equations we arbitrarily choose the value for
one unknowns, (or for two unknowns, in case of degeneration) and solve the
system (e.g. r1 = 0).

17
 18.4.2017.

35

MODI Method or a Modified Distribution

Method
• 2) Using the formula dij = cij – (ri + sj) we calculate the
differences for all unoccupied cells.
• The decision whether and which cell should be
occupied, as well as further procedure of changing
the inicial solution, is carried out on the same way as
it is described for the Stepping Stone method.
• These two optimization methods differ only in the
method of calculation of the difference dij
• To use MODI method it is necessary to know how to
use the Stepping Stone method as well.

36

Example - Optimization using MODI Method for

the Foster Generators transportation problem

• Again, we start with initial solution obtained with Vogel’s

aproximation method, so we have to check wether we
should occupy cells (1,3), (1,4), (2,2), (3,2), (3,3) and (3,4)
• For all occupied cells we setup equation ri+sj=cij:
• r1 + s1 = 3; r1 + s2 = 2; r2 + s1 = 7
• r2 + s3 = 2; r2 + s4 = 3; r3 + s1 = 2
• Here, we have 6 equations and 7 unknowns
• For example, let r1=0
• Then, r2=4, r3=-1, s1=3, s2=2, s3=-2, s4=-1

18
 18.4.2017.

37

• Using the formula dij = cij – (ri + sj) we calculate the

differences for all unoccupied cells.
• d1,3 = c13 – (r1 + s3) = 7-(0+(-2)) = 9
• d1,4 = c14 – (r1 + s4) = 6-(0+(-1)) = 7
• d2,2 = c22 – (r2 + s2) = 5-(4+2) = -1
• d3,2 = c32 – (r3 + s2) = 5-(-1+2) = 4
• d3,3 = c33 – (r3 + s3) = 4-(-1+(-2)) = 7
• d3,4 = c34 – (r3 + s4) = 5-(-1+(-1)) = 7
• Further procedure of changing the initial solution is
carried out on the same way as previously

38

The procedure of solving the transportation

problem
1. Check whether it is an open or a closed problem. If it is
an open problem add artifical (dummy) origin or
artifical (dummy) destination, and close it.
2. Find an initial feasible solution by employing the
required method. If the mehod is not strictly
determined, then use Vogel's aproximation method.
3. Conduct the optimization by employing the requiered
method. If the model suffers from the problem of
degeneration, in that case use MODI method

19
 18.4.2017.

39

The problem of degeneration in the

transportation problem
• Problem of degeneration may occur if we have more cells with the
same minimum values xij on "circular" movement in the process of
changing the initial solution, but can also occur at the very beginning (in
the initial decision)
• We will have the problem of degeneration if the number of occupied
cells is less than m+n-1
• We solve the problem with degeneration by introducing ε (epsilon) -
extremely small amount that we put on some unoccupied cell and from
it we can not subtract
• The position of setting ε is determined by the needs of closing the circle
in the process of changing the initial solution.
• We can test alternative positions for the ε
• The final solution can remain degenerated

40

Use of the software package QSB in solving

transportation problem
• Software package QSB, among other things, includes
a module for solving transportation problem
• The following slides shows how to use of the
software package QSB for solving transportation
problem

20
18.4.2017.

21
18.4.2017.

22
18.4.2017.

23
18.4.2017.

24