Optimization: Logistics Network Configuration

p1 c1
no limit w1 50,000
0 3

5 4
5

4 2

2 1
w2
p2 c2
2
60,000 100,000

customers
with demands
shown
c3
50,000
plants with
warehouses
capacities shown

Figure 1. A logistics network; unit transportation costs are given on each connecting
link. [Example 3-2 from Designing and Managing the Supply Chain by D. Simchi-Levi et al]

Goal: Find a plan of distribution to minimize total transportation cost, while satisfying
the demand at each customer location and the plant capacity limitations.

Heuristic #1: Customer chooses warehouse based on outbound

unit costs. Step 1: Each customer chooses warehouse with the
cheapest outbound-to-customer unit cost. Step 2: Route entire
demand for each customer through warehouse identified in Step 1.

What is the distribution plan, and what is the total cost?

1
 p1 c1
no limit w1 50,000
0 3

5 4
5

4 2

2 1
w2
p2 c2
2
60,000 100,000

customers
with demands
shown
c3
50,000
plants with
warehouses
capacities shown

Heuristic #2: Customer chooses warehouse based on inbound

and outbound unit costs. Step 1: For each customer, choose
warehouse with cheapest inbound-from-plant plus outbound-to-
customer unit costs. Step 2: Route entire demand for customer
through warehouse found in Step 1.

What is the distribution plan, and what is the total cost?

2
Exact solution of the Logistics Network Configuration Model

The problem of determining the optimal logistics distribution in Figure 1 is an example of

what is known as a linear program, and specifically in this context the “transhipment
problem.”

The problem described earlier can be framed as the following linear programming
problem. We have the following decision variables:

 x(p1,w1), x(p1,w2), x(p2,w1) and x(p2,w2) will be the flows from the plants to
the warehouses;

 x(w1,c1), x(w1,c2), x(w1,c3) will be the flows from the warehouse w1 to

customer zones c1, c2 and c3.

 x(w2,c1), x(w2,c2), x(w2,c3) will be the flows from warehouse w2 to customer

zones c1, c2 and c3

The problem we want to solve is:

minimize 0x(p1,w1) + 5x(p1,w2) + 4x(p2,w1)+ 2x(p2,w2) + 3x(w1,c1) + 4x(w1,c2) +

5x(w1,c3) + 2x(w2,c1) + 1x(w2,c2) + 2x(w2,c3)

subject to the following constraints:

x(p2,w1) + x(p2,w2)  60000 capacity, plant 2

x(p1,w1) + x(p2,w1) = x(w1,c1) + x(w1,c2) + x(w1,c3) conservation of flow, w1
x(p1,w2) + x(p2,w2) = x(w2,c1) + x(w2,c2) + x(w2,c3) conservation of flow, w2
x(w1,c1) + x(w2,c1) = 50000 demand, customer 1
x(w1,c2) + x(w2,c2) = 100000 demand, customer 2
x(w1,c3) + x(w2,c3) = 50000 demand, customer 3

all flows greater than or equal to zero.

To solve such a model to optimality, we can turn to an exact algorithm that is guaranteed
to find an optimal solution. It is not difficult to use EXCEL to represent this problem
within a spreadsheet (LPexample36.xls) and solve it using EXCEL’s Solver add-in.

First of all, we arrange the problem’s data: plant capacity, unit transportation costs, and
customer demands, in appropriate cells in a spreadsheet. In Figure 2, these data and their
labelling information appear in columns B through K and rows 5 through 9.

3
Figure 2. EXCEL model for Logistics Configuration Model in Figure 1.

Note that some of the terminology used by Solver is idiosyncratic. For instance,
variables, the unknowns in a model, are stored in cells designated as changing cells. In
Figure 2, the cells highlighted yellow contain the decision variables representing
shipments from plants to warehouses, while those highlighted in rose represent shipments
from warehouses to customers. The initial values of zero in each of these cells can later
be changed by the Solver.

The objective function to be minimized, total transportation cost, is stored in a “target

cell” highlighted blue. The location of the target cell, along with those of the changing
cells, must be input to the Solver via the Dialog box shown in Figure 3.

4
You can access the Solver Dialog box by going to “Tools” and selecting “Solver.” (If
Solver does not appear when you click Tools, go to ADD-INS and select Solver.)

Figure 3. The Solver Dialog box in EXCEL.

Along with references to the target cell and the changing cells, input to the Solver Dialog
box must include the constraints. The first line in the constraint section of the box, B15
<= B8, indicates that the function contained in cell B15, SUM(D15:E15) must not exceed
in value the contents of cell B8. In other words, the total amount shipped out of plant 2
cannot exceed its capacity, 60,000 units.

The second line indicates that the function contained in cell D16 must equal in value the
function contained in G14 and that in E15 must equal that in G15. Those constraints
ensure flow conservation: i.e., that the amount shipped into each warehouse exactly
balances the amount shipped out of it. Finally the third line contains three constraints,
each of which ensure that the correct amount is shipped to each customer.

Go to the spreadsheet now and investigate this model by clicking on the appropriate cells!

Now go to Tools invoke the Solver Dialog box, and click on “solve”. You should next see
the following screen:

5
Figure 4. An optimal solution to the transshipment model.

Figure 4 displays an optimal solution to the transhipment problem that calls for
production of 140,000 units at plant p1 and 60,000 (the specified limit) at plant p2. These
quantities appear in the yellow cells. As we see in the rose cells, customer c1 receives
50,000 units from warehouse w1, customer c2 receives 40,000 from warehouse w1 and
60,000 from w2, and customer c3 receives a shipment of 50,000 units from w1. The total
transportation cost of this solution is $740,000, as seen in the blue cell.

Note that the target cell utilizes EXCEL’s built-in function SUMPRODUCT, which, as
the name suggests, multiplies together corresponding elements in an array, and then sums
them up. For more information about SUMPRODUCT, Solver, or other features, use
EXCEL’s Help function.

The Sensitivity Report from EXCEL is shown in Table 1.

6
Adjustable Cells
Final Reduced Objective Allowable Allowable
Cell Name Value Cost Coefficient Increase Decrease
$D$14 p1 w1 140000 0 0 2 1
$E$14 p1 w2 0 2 5 1E+30 2
$D$15 p2 w1 0 5 4 1E+30 5
$E$15 p2 w2 60000 0 2 1 1E+30
$I$14 w1 c1 50000 0 3 2 1E+30
$J$14 w1 c2 40000 0 4 2 0
$K$14 w1 c3 50000 0 5 0 1E+30
$I$15 w2 c1 0 2 2 1E+30 2
$J$15 w2 c2 60000 0 1 0 2
$K$15 w2 c3 0 0 2 1E+30 0

Constraints
Final Shadow Constraint Allowable Allowable
Cell Name Value Price R.H. Side Increase Decrease
$B$15 changing cells shipped out 60000 -1 60000 40000 60000
$D$16 highlighted in yellow w1 140000 0 0 1E+30 140000
$E$16 highlighted in yellow w2 60000 3 0 60000 40000
$I$16 amt received c1 50000 3 50000 1E+30 50000
$J$16 amt received c2 100000 4 100000 1E+30 40000
$K$16 amt received c3 50000 5 50000 1E+30 50000

Table 1. Sensitivity Report from EXCEL, Logistics Network Configuration

problem.

A sensitivity report can help a manager answer “what if” questions. For example, what if
plant 1 could produce 1 additional unit of the product: how would that change our total
transportation cost? The “shadow price” of -1 reported in Table 1 tells us that for every
additional unit of capacity at plant 1, total cost would change by -1; i.e., decrease by $1.
This rate of decrease would hold for up an “allowable increase” of 60000 additional units
of capacity, according to Table 1.

We can answer a similar question: what if customer 2 demanded 1 additional unit of the
product? What would that do to total transportation cost? The shadow price of 4 means
that the total transportation cost would increase by $4, and this $4 per unit rate would
hold for up to 100000 additional units of demand.

7
 Design of a Logistics Network

An Integer Linear Model in EXCEL

An optimal solution to the transshipment (LPexample36.xls) linear programming model

specified the shipments from plants to warehouses and from warehouses to customers.
Though very valuable information, shipments are more properly considered operational
decisions rather than design decisions. True design decisions represent long term
commitments, such as choosing locations for plants and warehouses.

Can we build our own design decision support system? The answer has both good news
and bad news for us. Let’s save the bad news for later. The good news is that we can
indeed design networks using EXCEL! In fact, not only can we decide warehouse
locations, but plant locations as well. (We’ll assume that customer locations are beyond
our control.)

To illustrate, let’s consider Figure 1, augment the data, and run with it a bit. First we
consider p1 and p2 to be candidate plant locations, and add a third, p3 as well. That is,
these are potential locations for plants in a “clean paper” design from scratch – the actual
locations will be determined by solving an optimization model. Recall that there is only
one product in this scenario. Were there multiple products, we would consider “product
sourcing” decisions, that is, choosing a candidate location for each. We also consider
candidate warehouse locations w1, w2, and w3. We’ll assume that the firm wishes to
have no more than 2 plants and no more than 2 warehouses, although these kinds of
restrictions are not hard to remove. The following figure illustrates the new problem.

8
 p1 w1 c1
fc=35,000 50,000
0 fc= 20,000 3

5 4
3 5

4 2

2 1
w2
p2 2 fc =25,000 c2
2
fc= 40,000 100,000

2 1
1 2

3
w3
2
p3 fc= 30,000
fc =30,000 customers c3
with demands 50,000
candidate warehouses
candidate plants shown
with fixed costs
with fixed costs
wn
Figure 5. Graphical representation of the Network Design Example. Each candidate
plant would have a capacity of 100,000 units.

9
Figure 6. An EXCEL spreadsheet for the Logistics Network Design model.

Besides unit transportation costs between all the candidate locations, we need some more
pieces of data. Specifically, we need to know the annual fixed cost of opening, operating
and maintaining a facility, plant or warehouse, at each candidate location. This cost must
be on an annual basis to be consistent with the units of the original data. It would include,
for example, rent or amortized construction costs. These costs appear in the cells beneath
the heading “fixed cost” in the EXCEL spreadsheet (MIP1.xls) in Figure 6.

We also need an unknown to represent each plant location decision, and each warehouse
location decision. These are found beneath the heading “Selection” in changing cells B4
through B6, and changing cells B21 through B24, respectively. These changing cells are
of a special type known as binary valued, since a location decision is yes/no, with no in-
betweens. A “yes” for a candidate location indicates the location should be included in
the final solution, while “no” means the location should be excluded. In our formulation,
“yes” is represented by the value 1, and a “no” by 0.

Recall that the target cell, H33 (highlighted in blue), contains a formula for the total cost,
which is to be minimized. Now not only does this include the costs of transporting all
shipments, but the sum of fixed costs associated with the plants and warehouses chosen.
This total cost is contained in the EXCEL formula
=SUMPRODUCT(G4:I6,E12:G14)+SUMPRODUCT(E21:G23,E28:G30)+SUMPRODU
CT(C4:C6,B4:B6)+SUMPRODUCT(C21:C23,B21:B23)
The first two SUMPRODUCTS contain transportation costs, while the second two
contain the fixed costs of plants and warehouses, respectively.

10
The constraint system is more complicated than that of the transshipment problem as
well. The 20 constraints in the model are displayed and annotated in Table 2.

Constraint Comment
1 $E$15=$C$28 amount received from all plants at w1 equals amount
shipped out from w1 to all customers
2 $F$15=$C$29 amount received from all plants at w2 equals amount
shipped out from w2 to all customers
3 $G$15=$C$30 amount received from all plants at w3 equals amount
shipped out from w3 to all customers
4 $E$31=$E$24 amount received from all warehouses
by customer c1 equals demand at c1
5 $F$31=$F$24 amount received from all warehouses
by customer c2 equals demand at c2
6 $G$31=$G$24 amount received from all warehouses
by customer c3 equals demand at c3
7 $C$12<=$E$4 forcing constraint: can’t ship product out of plant p1 in
any amount, unless location 1 is selected
8 $C$13<=$E$5 forcing constraint: can’t ship product out of plant p2 in
any amount, unless location 2 is selected
9 $C$14<=$E$6 forcing constraint: can’t ship product out of plant p3 in
any amount, unless location 3 is selected
10 $B$7<=$B$8 no more than 2 plants can be chosen
11 $E$15<=$A$21 forcing constraint: can’t receive product into warehouse
w1 in any amount, unless location 1 is selected
12 $F$15<=$A$22 forcing constraint: can’t receive product into warehouse
w2 in any amount, unless location 2 is selected
13 $G$15<=$A$23 forcing constraint: can’t receive product into warehouse
w3 in any amount, unless location 3 is selected
14 $B$24<=$B$25 no more than 2 warehouses can be chosen
15 $B$4=binary location selection is a binary (0,1) variable
16 $B$5=binary location selection is a binary (0,1) variable
17 $B$6=binary location selection is a binary (0,1) variable
18 $B$21=binary location selection is a binary (0,1) variable
19 $B$22=binary location selection is a binary (0,1) variable
20 $B$23=binary location selection is a binary (0,1) variable
Table 2. Constraints for the network design model in Figure 6.

Constraints 1-6 correspond to those we encountered in the transshipment model. The

other constraints are necessitated by the location decisions for both plants and
warehouses. The logical forcing constraints 7-10 ensure that no shipment is promised
from a location without a plant! Likewise 11-13 are logical forcing constraints that ensure
that no shipment goes to a location without a warehouse. Constraints 10 and 14 place
limits on the number of plants and warehouses to be selected, respectively. Finally,
constraints 15-20 tell the Solver that the changing cells in the specified locations can only
take on the values of 0 and 1.

11
Once again we go to the Solver dialog box and click “solve” resulting in the optimal
solution seen in Figure 7.

Figure 7. An optimal solution to the Logistics Network Design example problem

determined by the EXCEL Solver.

12
 p1 w1 c1
35,000 50,000
20,000

50,000 100,000
w2
p2 25,000
c2
50,000 50,000
40,000 100,000

50,000
100,000
w3
p3 30,000
30,000 customers c3
with demands 50,000
candidate warehouses
candidate plants shown
with fixed costs
with fixed costs
shown
Figure 8. Graphical representation of the optimal network design found by EXCEL.
Plants are located at p2 and p3, warehouses at w2 and w3. Shipments are shown by
thick red arrows. The total cost is $675,000.

Now for the bad news: this design model, known as a mixed integer program, is a much
harder problem in terms of the amount of computation required, than the transshipment
linear program. In fact, there is no guarantee that any optimization software would not
have to solve as many linear programs as there are combinations of plant and warehouse
locations. This is not of much consequence when there are only 3 candidate locations for
each, as in our illustration. However, suppose we were tackling a larger design instance
with 15 candidate warehouse locations, and we wished to select exactly than 7 of them.
There are 6435 different ways of selecting 7 of 15 locations, and we might have to solve
virtually as many linear programming models, if we were “unlucky.” Were we to select
15 of 30 candidate locations, the number of combinations would jump beyond
15,000,000, a phenomenon known as the “combinatorial explosion”! In these cases,
decision makers often must turn to heuristic methods. Such methods attempt to provide
good answers with a reasonable level of computation, but the answers are not guaranteed
to be optimal.

13