Scribd
Upload
224 views

OD5 PL Dynamic Programming A

The document discusses two examples of using dynamic programming to solve optimization problems. The first is a deterministic discrete problem about minimizing production costs over three weeks. The second is a deterministic continuous problem about maximizing a function subject to a constraint.

Uploaded by

carolinarvsocn
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
224 views

OD5 PL Dynamic Programming A

The document discusses two examples of using dynamic programming to solve optimization problems. The first is a deterministic discrete problem about minimizing production costs over three weeks. The second is a deterministic continuous problem about maximizing a function subject to a constraint.

Uploaded by

carolinarvsocn
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

OPTIMIZAO E DECISO 10/11

PL #5 Dynamic Programming
Alexandra Moutinho
th

(from Hillier & Lieberman Introduction to Operations Research, 8 edition)

DYNAMIC PROGRAMMING DETERMINISTIC DISCRETE PROBLEM


The Build-Em-Fast Company has agreed to supply its best customer with three widgets (small
mechanical device) during each of the next 3 weeks, even though producing them will require some
overtime work. The relevant production data are as follows:
Week
1
2
3

Maximum Production, Maximum Production, Production Cost per Unit


Regular Time (rn)
Overtime (mn - rn)
Regular Time (cn)
2
2
$300
3
2
$500
1
2
$400

The cost per unit produced with overtime for each week is $100 more than for regular time. The cost
of storage is $50 per unit for each week it is stored. There is already an inventory of two widgets on
hand currently, but the company does not want to retain any widgets in inventory after the 3 weeks.
Management wants to know how many units should be produced in each week to minimize the
total cost of meeting the delivery schedule.
Solve this problem using dynamic programming.
Resolution:
The decisions that need to be made are the number of units to produce each week, so the
decision variables are
= number of widgets to produce in week n, for n = 1, 2, 3.

To choose the value of xn, it is necessary to know the number of widgets already on hand at the
beginning of week n. Therefore, the state of the system then is
= number of widgets on hand at the beginning of week n, for n = 1, 2, 3.

Because of the shipment of three widgets to the customer each week, note that
3.

Also note that two widgets are on hand at the beginning of week 1, so
= 2.

To introduce symbols for the data given in the above table for each week n, let
= unit production cost in regular time,
= maximum regular time production,
= maximum total production.

The corresponding data are given in the following table.

n
1
2
3

rn
2
3
1

mn
4
5
3

cn
300
500
400

We wish to minimize the total cost of meeting the delivery schedule, so our measure of
performance is total cost. For n = 1, 2, 3, let
(

Thus,

) = cost incurred in week n when the state is


(

and the decision is

) + 50 max (0,

+ 100 max(0,

Using the notation in Sec. 10.2, let

3)

( ) = optimal cost for week n onward (through the end of week 3) when starting week
n in state sn, for n = 1, 2, 3.

Given , the feasible values of are the integers such that


and
(assuming
3) in order to provide the customer with 3 widgets in week n. Thus, because
+
3, the optimal value of
(denoted by ) is obtained from the following recursive
relationship:
(

( ) = min

where

( ) = 0 for

3)],

= 0.

Recall that the company does not want to retain any widgets in inventory after the three weeks,
so = 0. Therefore, the optimal policy for week 3 obviously is to produce just enough widgets to
have a total of three to ship to the customer, so
=3

Given and
), we can use the recursive relationship to solve for the optimal policy for week
2 and then for week 1. These calculations are shown below.
For n = 3:

s3
0
1
2
3 s3

f3 (s3)
1,400 (1)
900 (2)
400 (3)
0

x3
3
2
1
0

(1)

1400 = 1 400 + 2 (400 + 100): cost of producing 3 units in week 3;


900 = 1 400 + 1 (400 + 100): cost of producing 2 units in week 3;
(3)
400 = 1 400 + 0 (400 + 100): cost of producing 1 unit in week 3.
(2)

For n = 2:

x2

s2
0
1
2
3(2)

(+)
(+)
(+)

1,400

f2(s2, x2) = p2(s2, x2)+ f3*(s2+ x2-3)


1
2
3
4
(+)
(+)
(3)
2,900
3,050(4)
(+)
2,400 2,450
2,600
1,900 1,950 2,000
2,250
1,450 1,500 1,650 2,300(*)

5(1)
3,200(5)
2,850
2,900(*)
2,950(*)

*
f2 (s2)

x2

2,900
2,400
1,900
1,400

3
2
1
0

(1)

Maximum production at week 2: 3+2 = 5 (


)
Maximum stock at beginning of week 2: 2 (s1)+4 (m1)-3 = 3
(3)
(0,3)
(0,3) + (0 + 3 3) = 500 3 + 100 max(0,3
3) + 1400 = 2900
(2)

Optimizao e Deciso 09/10 - PL #5 Dynamic Programming - Alexandra Moutinho

3) + 50 max(0,0 + 3
2

(4)

(0,4)
(0,4) + (0 + 4 3) = 500 4 + 100 max(0,4 3) + 50 max(0,0 + 4
3) + 900 = 3050
(5)
(0,5)
(0,5) + (0 + 5 3) = 500 5 + 100 max(0,5 3) + 50 max(0,0 + 5
3) + 400 = 3200
(+)
Options that do not assure minimum delivery.
(*)
Are these values necessary? No, because they imply that
0 since
3 > 3.

For n = 1:

x1

s1
2

f1(s1, x1) = p1(s1, x1)+ f2*(s1+ x1-3)


0
1
2
3
4
3,200 3,050 3,000 2,950

*
f1 (s1)

x1

2,950

Hence, the optimal plan is to produce four widgets in period 1, store three of them until period 2,
and produce three in period 3, with a total cost of $2,950.
(has 2)

=4
Produce 4
Deliver 3
Store 3

=0
Produce 0
Deliver 3
Store 0

=3
Produce 3
Deliver 3
Store 0 ( = 0)

Total cost
$2,950

DYNAMIC PROGRAMMING DETERMINISTIC CONTINUOUS PROBLEM


Consider the following nonlinear programming problem:
Maximize
subject to

2. (There are no nonnegativity constraints.)

Use dynamic programming to solve this problem.


Resolution:

Let
be the decision variable at stage 1 and
be the decision variable at stage 2. The key to
applying dynamic programming to this problem is to interpret the right-hand side of the constraint
as the amount of a resource being made available to activities 1 and 2 (whose levels are and ).
Then the state of the system entering stage 1 (before choosing
and ) and entering stage 2
(before choosing ) is the amount of the resource still available for allocation to the remaining
activities. Consequently, letting denote the state of the system entering stage , we have
= 2 and

For

= 2

At stage 2, we solve the following problem with variable


Maximize
subject to

)=

The optimal solution is

,
, with

)=

= :
For

At stage 1, we solve the following problem with variable


Maximize
subject to

)
= 2.

( ) <=>

(2,

( )
:
2

Computing the first partial derivative to determine maxima and minima:

Optimizao e Deciso 09/10 - PL #5 Dynamic Programming - Alexandra Moutinho

=2

+4

= 0 =>

1, 0, 1

Computing the second partial derivative to determine which are maxima and minima:
(

Hence,

12

(2,

+4

< 0 for
> 0 for

1, 1 => local maxima


= 0 => local minimum

) is locally concave around

1 and

= 1.

(2, 1)
(2, 1) = 1, whereas
Since
2, 2 = 0 at the endpoints of the feasible region
(
2), we have both
1 and = 1 as global maximizers and (2) = 1.
= :

Hence, the optimal solutions are


function value of = 1.

( )
1

) = ( 1, 1), and

1, 1

) = (1, 1) with an optimal objective

Optimizao e Deciso 09/10 - PL #5 Dynamic Programming - Alexandra Moutinho

You might also like