1 / 32

# Chapter 6 - PowerPoint PPT Presentation

Chapter 6. Toward Proper Modeling. Mathematical Modeling. Part art, part science Considerable room for judgment Assumptions concerning variables, constraints, coefficients Assumptions affect results How do you know if the model is “right”?. Structural Component Identification.

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.

## PowerPoint Slideshow about ' Chapter 6' - milla

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

### Chapter 6

Toward Proper Modeling

• Part art, part science

• Considerable room for judgment

• Assumptions concerning variables, constraints, coefficients

• Assumptions affect results

• How do you know if the model is “right”?

• Variable identification

• Objective function specification

• Constraint set-up

• Model structure

• Variables may be included for technical, accounting or convenience reasons

• The variables are the “unknowns.” LP solution determines the level of the variables.

• Technical variables are the true unknowns.

• Accounting variables normally summarize some other activities in the model.

• Convenience variables may help simplify the model structure or serve some similar purpose.

There are cases when multiple variables must be defined for what may, at first, seem to be a single activity.

• (a) When more than one process can be used to produce the same output using different resource mixes; e.g., the production of an item using either of two different machines.

• (b) When different processes produce different outputs using common resources; i.e., one can use essentially the same resources to produce either 2 x 4 or 4 x 4 sawn lumber.

• (c) When products can be used in several ways; e.g., selling chickens that can be quartered or halved.

• We have seen that properly defining variables is essential in many types of common LP problems.

Criteria may also be developed where two variables may be treated as one. The simplest case occurs when the coefficients of one variable are simple multiples of another

(aij = Kaim and cj = Kcm).

The second case occurs when one variable uniquely determines another; i.e., when n units of the first variable always implies exactly m units of the second.

• Solution is only useful if the objective function is correctly specified.

• Identifies the optimal point.

• Multiple objective models are discussed in the multi-objective and risk chapters.

• Range analysis can be used to see for what values of the objective function the currently solution holds.

• Technical constraints depict limited resources, intermediate products, or contractual requirements. Technical constraints may also express complementary, supplementary, and competitive relationships among variables.

• Institutional constraints reflect external regulations imposed on the problem. Examples include credit limits or farm program participation requirements.

• Subjective constraints are imposed by the decision maker or

• modeler. These might include a hired labor limitation based on the decision maker's willingness to

• supervise labor.

• Convenience constraints facilitate model interpretation and may be included to sum items of interest.

• Model formulation constraints aid in problem depiction. These include constraints used in conjunction with approximations.

• Heady and Candler argue that multiple constraints are needed to depict availability of a resource whenever the marginal rate of factor substitution between resource usage in different time periods does not equal one. (Labor is an example – can’t substitute July labor for June labor.)

• Constraints must be developed so that the resources available within a particular constraint are freely substitutable. Cases of imperfect substitution will require multiple constraints.

• An LP solution will include no more variables at a nonzero level than the number of constraints (including the number of upper and lower bounds). Thus, the number of constraints directly influences the number of nonzero variables in the optimal solution.

• Subjective constraints should not be imposed before determining their necessity. Often, subjective constraints "correct" model deficiencies. But the cause of these deficiencies is frequently missing either technical constraints or omitted variables.

• LP model constraints have higher precedence than the objective function. The first major effort by any LP solver is the discovery of a feasible solution.

• The modeler must question whether a constraint should be established so it always restricts the values of the decision variables. Often, it may be desirable to relax a constraint allowing resource purchases if the value of a resource becomes excessively high.

• Modelers should be careful in the usage of minimum requirement constraints. Minimum requirements must be met before profit seeking production can proceed.

• Third, judicious use should be made of equality constraints. Modelers should use the weakest form of a constraint possible.

Consider the following example:

Max 3X + 2Y

s.t. X - Y ? 0

X LE 10

Y LE 15

Where ? is the constraint type, X depicts sales and Y production. Suppose we have made a mistake and have specified the cost of production as a revenue item (i.e., the +2Y should be -2Y in the objective function).

If the relation is an equality, then the optimal solution is X = Y = 10 and we do not discover the error. On the other hand, if the relation is LE then we would produce Y = 15 units while selling only X = 10 units and we would see the error.

• Model building is an iterative process.

• People work in different ways.

• I break a large model into smaller pieces, develop those, and then put them together.

• Another alternative is to follow the steps laid out in McCarl and Spreen.

Development of Model Structure

Example: A profit maximizing firm produces 4 crops s.t. land and labor constraints. Crops are grown at different times of the year.

Crop1 is planted in the spring and harvested in the summer. Crop2-3 are planted in the spring and harvested in the fall. Crop4 is planted following crop1 and harvested in the fall.

1. Make a table laying out potential variables

across the top and constraints /objective function down the side.

2. Enter profits for crops and resource coefficient and endowment.

• Reading down the column for a variable – these coefficients represent the simultaneous requirements for this variable.

• Choice is modeled across variables, never within a variable. If there are two production alternatives, there must be two variables. If there are two possible uses for a decision item, there will need to be two variables (example selling tables or using tables to make sets).

• Resources within a constraint are homogeneous. If there are different classes of the resource (e.g. good land and not-so-good land), you will need separate constraints.

Rules

• 1. All coefficients in a row have common numerators – e.g. operating on the same units (consistently in lbs or consistently in kilos, not some of each).

• All coefficients in a column have common denominators (e.g. per unit of X produced).

• A chicken is cut into four parts: breasts, legs, necks, and giblets.

• Each chicken weighs 3 pounds.

• There are 1,500 chickens available.

Last constraints is 1/3 Y because chickens weigh 3 pounds. Equal to

Y Le 4500 (total pounds), which would probably be the way I would write it.

Here, we are trying to disassemble across a row. Look at what happens

if we set Y=1. If we have Y equal to 1, then X1 (breast quarters in lbs) can

equal up to 2 pounds, with all else zero. Alternatively, we could get about 3

pounds of leg quarters from this one pound of chicken, or10 pounds of necks.

All would be feasible in this formulation.

Now let us be able to sell mixed quarter packs, which

is an arbitrary combination of legs and breasts.

And let us be able to debone the chicken and sell

the meat for \$1.20 a pound.

Here, instead of having one use for some items,

there are various uses – selling them as is, deboning, or

putting them in packs. These alternative uses

require new variables.

This specification allows you to use BQ’s and LQ’s twice,

once as direct sales or meat, the other as part of

a quarter pack.

• Consider labor

• It is paid one normal hourly rate and another “overtime” rate

• Suppose a firm has 77 regular hours of labor at \$10/hour and up to 27 more hours at \$15 per hour.

• Assume the firm manufactures chairs, which take 10 hours of labor each, with an average of 7 hours regular labor and 3 hours overtime.

• Chairs sell for \$220 each.

It is easy to see why this is wrong. Even if only

one chair is produced, the company has to hire

overtime. Also, the overtime limit becomes binding

at 9 chairs, when in fact more chairs could be

produced using “regular” labor.

In this specification, the cheaper labor would

be used first.

It can be useful to set the model (either in

your mind or with a special restriction) to

one unit of a decision variable and

check that all the inputs and other

relationships make sense before running

the model “for real.”

• Time frame: defines characteristics of data. All data must be consistent.

• Uncertainty: Coefficients are almost never known with certainty

• Data Sources: May come from statistical estimation or from deductive processes (economic engineering approach)

• Deductive Approach

• We compute profit per unit by taking yield*price and subtracting costs

• An art – must always defend the assumptions in published work. Why that yield? Why that price?

• Sensitivity can be useful to test if results are robust

• Used to be costly to add rows and columns to an LP.

• Parsimonious model building was thus necessary.

• Computational time is now cheap.

• Rows and columns can be added to save the modeler time in interpretation of results.

• For example, total balance of loans each period is an accounting identity which the modeler could calculate based on decision variables (starting balance + new loans – principal paid), but it is easier to put in a transfer row and let the computer do it for you.

• The raw computer output is almost never very useful to anyone other than the modeler.

• Tables should be developed that make the model results clear to others.

• Sensitivity analysis (changing key assumptions) is often needed for journal papers and must be reported as well.