Multiple-Objective (Goal) Programming Model for Feed Formulation:
An Example for Reducing Nutrient Variation
F. Zhang and W. B. Roush1
Department of Poultry Science, The Pennsylvania State University, University Park, Pennsylvania 16802-3501
ABSTRACT A multiple-objective programming (MOP)
model was applied to the feed formulation process with
the objectives of minimizing nutrient variance and min-
imizing ration cost. A MOP model was constructed for
a broiler grower ration (3 to 6 wk) and formulated with
a Microsoft Excel威 solver. Twenty-one ingredients with
17 nutrients were included in the formulation. Amino
acids were based on digestible values. The following ob-
jectives were considered as soft constraints: (1) meeting
the nutrient requirements; (2) meeting the ingredient re-
strictions; and (3) meeting nutrient ratios, including cal-
cium to phosphorus and the relationship of amino acids
to lysine (ideal amino acid ratios). Hard constraints con-
sidered were (1) a least-cost ration and (2) minimal nutri-
(Key words: multiple-objective goal programming, ration variance reduction, ration formulation)
INTRODUCTION
Traditional linear programming (LP) for feed formula-
tion is used for problems with a single goal, which is
usually to minimize the cost of the ration. It is not unusual
for a nutritionist to have other feed formulation goals,
such as minimizing the nutrient variance, in addition to
reducing the cost of the rations. However, multiple objec-
tives, although desirable, can be conflicting. Thus, a tradi-
tional LP application to a multiple-objective problem may
result in an infeasible solution. A multiple-objective pro-
gramming (MOP) model is a flexible alternative to the
traditional LP approach. The MOP does this by making
inequalities into hard constraints and including deviation
variables (slack and surplus variables) in the objective
equation. Therefore the deviation variables are the object
for optimization. Thus, the MOP model provides the solu-
tion of multiple goals by allowing acceptable solutions
for conflicting objectives (as opposed to optimal solutions;
Ignizio and Cavalier, 1994).
The use of MOP models is relatively recent. MOP models
began their development in the 1950s; however, they have
2002 Poultry Science Association, Inc.
Received for publication June 8, 2001.
Accepted for publication September 27, 2001.
1
To whom correspondence should be addressed: wbr@psu.edu.
182
ent variances for protein, methionine, and lysine. It was
found that (1) the MOP model was more flexible in pro-
viding a compromise solution than a traditional feed for-
mulation with a linear program, (2) the MOP model was
Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019
able to handle several conflicting objectives simultane-
ously as compared to the traditional linear programming
approach that could handle only one objective, and (3)
the MOP model gave the best compromise solution that
would satisfy multiple decision makers when trade-offs
were made between the ration cost and minimum vari-
ances of protein and methionine. The MOP model is an
efficient tool to assist the decision-making process
through solving a series of linear/nonlinear programs
and by interacting with decision-makers.
2002 Poultry Science 81:182–192
only received attention since the mid-1970s and have sig-
nificantly matured in the 1990s (Mollaghasemi and Pet-
Edwards, 1997). MOP models have been applied to mixing
and blending problems (Satoris and Spruill, 1974; Steurer,
1984), solving dairy ration formulations (Lara, 1993, 1996;
Lara and Romero, 1992, 1994), dairy nutrient management
(Tozer and Stokes, 2001), and land-use problems (Zander
and Kächele, 1999).
The process of MOP involves defining goal priorities
and then iteratively finding solutions of linear or nonlinear
programs. Each goal is solved sequentially according to
its priority, with the previous goal or goals held constant
in the model. The importance of objectives in feed formula-
tion can be distinguished by ordinal (absolute) or cardinal
(weighted) rankings. There can be several trade-offs
among multiple, conflicting feed formulation objectives.
The solutions of interest are the set of efficient solutions
in which a better solution cannot be found that would
improve any of the objectives without sacrificing at least
one of the objectives. Further information on MOP or goal
programming methods can be found in Winston (1994)
and Oberstone (1997).
The objective of this research was to apply a MOP model
to the goals of minimizing protein, methionine, and lysine
variation and formulating a least-cost ration.
Abbreviation Key: MOP = multiple-objective (goal) programming.
 MULTIPLE-OBJECTIVE PROGRAMMING FOR FEED FORMULATION
METHODS AND PROCEDURES
Feed Ingredient Data
As an example, a 3-to-6-wk broiler ration was formu-
lated from nutritional data provided by a Dutch commer-
cial feed manufacturer. The nutrient composition is listed
in Table 1, in which there are 21 ingredients and 17 nutri-
ents. The amino acid levels in feed ingredients were based
on digestible values. Nutrient requirements for the 3-to-6-
wk broiler ration are listed in Table 2. The ingredient prices,
restrictions, and nutrient variances are listed in Table 3.
The vitamin and mineral requirements were considered
as fixed percentages in the ration.
MOP Model
The MOP feed formulation model is listed in Figure 1.
When working with several goals, conflicts can also result
in an infeasible solution with a traditional LP solution.
MOP allows a compromise solution that overcomes the
problem of infeasibility that can occur in linear programs.
The nutritionist may encounter problematic constraints
(nutrient requirements and ingredient restrictions) when
some goals may not be met in some MOP feed formulation
models. However, in this MOP model, the goals were
treated as real constraints, as the requirements could be
satisfied using current ingredients. The nutrient ratio re-
quirements were to meet the ideal amino acid ratios in the
ration (Emmert and Baker, 1997).
The first objective (ration cost with ranking of P1) was
a conventional least-cost feed formulation objective. The
second objective (nutrient variances with ranking of P2)
was included to minimize the variances of protein, lysine,
and methionine so that a ration with high nutrient quality
could be attained. The w1 to w3 are the weights of each
goal within the second objective. The rankings and weights
could be ordinal or cardinal numbers according to the
importance determined by the decision-maker. Parentheti-
cally, goal weights are reflective of and subject to the im-
portance the decision-maker places on a particular variable
in relationship to the other variables. There is not a correct
level or set of weights. Under another circumstance, the
goal weights may be different because of a change of priori-
ties by the decision-maker. Note that there is not a numeri-
cal value assigned to Pk. Pk refers to the order (priority)
of the solution sequence, which, again, is subject to the
decision-maker. Microsoft Excel威 Solver (Office 97, Version
8.) was used to solve the problem.
RESULTS AND DISCUSSION
Efficient Solutions for the MOP Model
In the MOP model, only two objectives were considered
(minimizing the ration cost and nutrient variances of pro-
tein, lysine, and methionine). An efficient solution means
that no other solution can improve any of the objectives
without affecting at least one other objective. The feasible
183
region was a convex set formed by hard constraints. In
order to find efficient solutions, it was necessary to solve
the model four times. The process was as follows:
1. To set the first objective, minimization of ration cost,
with a higher priority than the second objective
2. To set the second objective, minimization of protein
variance, with a higher priority than the first objective,
and set w1 larger than w2 and w3 (e.g., w1 = 1, w2 =
w3 = 0)
3. To set the second objective, minimize lysine variance,
with a higher priority than the first objective, and set
w2 much larger than w1 and w3 (e.g., w2 = 1, w1 = w3
= 0)
4. To set the second objective, minimize methionine vari-
Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019
ance, with a higher priority than the first objective,
and set w3 larger than w1 and w2 (e.g., w3 = 1, w1 =
w2 = 0).
The results of the above process are listed in Table 4. If
the variances of protein, lysine, and methionine are not
minimized, the lowest ration cost was 39.34 Dfl/cwt. When
the variance of protein, lysine, and methionine were mini-
mized individually, the ration cost was 43.65, 42.14 and
42.98 Dfl/cwt, respectively.
A comparison of the calculated nutrient means for the
rations (Table 4) indicated that the ration with the lowest
cost resulted in a higher protein level than the other rations.
In general, protein is a costly nutrient. In this ration, how-
ever, the protein became less expensive because of the
request for specific levels of amino acids and minimal
variances of protein, methionine, and lysine. The rations
with a minimized variance for protein had higher calcu-
lated methionine mean levels (0.52% compared to 0.48%),
and the rations with minimized summation of the vari-
ances of protein, methionine, and lysine also had higher
calculated methionine mean levels (0.52% compared to
0.48%).
Other nutrients had similar calculated values for the
five rations. Rations with minimized variance (i.e., protein,
lysine, and methionine or for the summation of the three
variances) had higher percentages of added DL-methio-
nine as compared to the ration with minimized cost, which
may explain why the rations with minimized variance (for
protein, lysine, or methionine) had higher costs.
Ingredient percentages in the five rations had different
patterns. The percentage of corn gluten meal was lower
in the rations with minimized variance(s) as compared to
the ration with minimized ration cost. The lower inclusion
of corn gluten meal is understandable, because corn gluten
meal is a highly variable ingredient. The percentage of
corn gluten meal used in rations was limited to 5 to 10%.
Ingredients not selected for the ration with minimized cost
had been included in the four rations at different per-
centages.
It is beneficial to include as many ingredients as possible
when formulating a ration, because the complementary
nature of nutrients in different ingredients promotes the
reduction of nutrient variation. To illustrate, Ration I in
Table 5 used two ingredients, corn and soy, which have
low standard deviations for protein. The ingredient peas,
 184

TABLE 1. Nutrient composition (nutrient means)1 for the multiple-objective (goal) program variance reduction problem

Nutrient ME CP Fat Fiber Lys Met Met + Cys Thr Trp Starch LA2 Ca TP2 AP2 Na Cl K

kcal/kg (%)
Corn 3,272 8.7 3.8 2.1 0.14 0.14 0.28 2.32 0.42 61.9 1.88 0.03 0.29 0.02 0.01 0.05 0.36
Wheat meal 3,012 11.9 1.7 2.3 0.29 0.17 0.40 0.28 0.10 58.6 0.68 0.07 0.34 0.08 0.01 0.04 0.42
Dutch peas 2,716 22.7 1.1 5.3 1.35 0.20 0.44 0.68 0.10 42.6 0.40 0.06 0.36 0.14 0.01 0.06 1.10
Tapioca meal 2,976 2.5 0.4 3.7 0.00 0.00 0.00 0.00 0.00 69.3 0.00 0.02 0.03 0.02 0.01 0.04 0.73
Rapeseed meal 1,801 39.2 1.5 10.7 1.56 0.68 1.37 1.25 0.38 5.9 0.21 0.76 1.30 0.19 0.02 0.04 1.31
Fish meal 3,356 70.6 9.4 0.0 4.62 1.90 2.47 2.55 0.68 0.0 0.08 2.64 2.09 2.09 1.03 1.62 1.39
Meat meal 3,210 58.3 14.0 2.6 2.63 0.67 1.03 1.64 0.41 0.0 0.78 4.91 2.45 2.45 0.88 0.88 0.60
Feather meal 3,172 82.4 7.6 0.0 1.32 0.44 3.49 3.04 0.37 0.0 0.53 0.46 0.24 0.24 0.08 0.13 0.10
Vegetable oil 8,952 0.0 99.5 0.0 0.00 0.00 0.00 0.00 0.00 0.0 45.00 0.00 0.00 0.00 0.00 0.00 0.00
Sunflower seed meal 1,445 31.9 1.9 22.3 0.84 0.60 1.02 0.87 0.36 3.4 0.80 0.37 1.06 0.10 0.04 0.12 1.41
Wheat middlings 1,822 45.6 3.5 8.6 0.44 0.18 0.44 0.34 0.17 19.9 1.40 0.76 1.30 0.19 0.03 0.30 1.37
Heated soybean 3,451 35.6 18.9 5.3 1.93 0.44 0.90 1.16 0.39 4.8 9.70 0.23 0.54 0.13 0.01 0.03 1.76
Soybean meal 2,185 46.7 1.8 3.7 2.63 0.61 1.23 1.58 0.54 5.4 0.63 0.27 0.64 0.16 0.02 0.02 2.23
Corn gluten meal 3,599 60.7 3.7 1.1 0.92 1.61 2.82 1.90 0.34 16.7 1.63 0.04 0.46 0.13 0.01 0.05 0.16
Alfalfa meal 970 16.9 2.7 26.5 0.47 0.15 0.44 0.42 0.16 3.0 0.41 1.85 0.26 0.24 0.10 0.49 3.08
ZHANG AND ROUSH

Animal fat 8,471 0.0 99.3 0.0 0.00 0.00 0.00 0.00 0.00 0.0 8.94 0.00 0.00 0.00 0.04 0.00 0.02
Salt 0 0.0 0.0 0.0 0.00 0.00 0.00 0.00 0.00 0.0 0.00 0.00 0.00 0.00 38.60 60.00 0.00
Limestone 0 0.0 0.0 0.0 0.00 0.00 0.00 0.00 0.00 0.0 0.00 39.00 0.00 0.00 0.00 0.00 0.00
Dicalcium phosphate 0 0.0 0.0 0.0 0.00 0.00 0.00 0.00 0.00 0.0 0.00 21.06 20.05 20.50 0.00 0.00 0.00
DL-Met 0 58.6 0.0 0.0 0.00 98.00 98.00 0.00 0.00 0.0 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Lys 0 95.8 0.0 0.0 78.00 0.00 0.00 0.00 0.00 0.0 0.00 0.00 0.00 0.00 0.00 0.00 0.00
1
Data from a commercial Dutch feed manufacturer, 1997 (personal communication).
2
LA = Linoleic acid; TP = total P; AP = available P.

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

 MULTIPLE-OBJECTIVE PROGRAMMING FOR FEED FORMULATION 185
1
TABLE 2. Nutrient requirements used in the multiple-objective ingredient percentage must sum to one. By careful calcula-
(goal) program variance reduction problem
tion of the variance (which is the optimization task of a
Requirements nonlinear program), the variance, which is the standard
Nutrient Unit Minimum Maximum deviation squared, will be decreased because numbers less
than 1 will be smaller after they are squared.
ME kcal/kg 3,100.00 3,300.00
Protein % 20.00 21.00
Ether extract (fat) % 0.00 11.00 Examining the Trade-off Between
Fiber % 0.00 3.75
Lys % 1.07 1.50
Least-Cost and Minimal Variances
Met % 0.43 1.00
Met + Cys % 0.79 1.70 Feed quality depends partly on nutrient variation in a
Thr % 0.60 1.00 ration, with the influence of variation on the probabilities
Trp % 0.19 0.50 of meeting the nutrient requirements of animals. Usually
Starch % 36.00 50.00
Linoleic acid % 1.00 2.50 a smaller variation improves the feed quality of balanced
nutrients. McCoy et al. (1994) have shown that diet unifor-

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

Ca % 0.55 0.71
Total P % 0.35 0.55 mity, through variance reduction, improves broiler perfor-
Available P % 0.30 0.40
Na % 0.14 0.16 mance. In addition, lower variability of nutrients reduces
Cl % 0.15 0.40 the possibility of nutrient waste.
K % 0.10 1.00 From Table 4, it was noted that minimal nutrient vari-
1
Data from an anonymous commercial Dutch manufacturer, 1997 ance for protein, lysine, or methionine would lead to high
(personal communication). ration cost (from 42.14 to 43.65 Dfl/cwt). How is it possible
to make a trade-off between least cost and minimizing the
nutrient variance?
which has a lower protein level than soy but a higher
standard deviation than soy, was used in Ration II. Rations Minimizing Ration Cost Without
I and II had the same protein level. In Ration II, however,
the standard deviation for protein was lower than that of
Regard to Nutrient Variance
Ration I. In the previous section, it was noted that the least cost
In general, nutrient variance has the following formula: price was 39.34 Dfl/cwt. The lowest standard deviation
n for protein was 0.5395. The lowest standard deviation for
∑vijx2i , lysine was 0.0121, and the lowest standard deviation for
methionine was 0.0043. If a ration included all four charac-
i=1
teristics, it would be the ideal ration. However, these are
where vij is jth nutrient variance of ith ingredient, and xi conflicting objectives that would result in an infeasible
equals the percentage of the ith ingredient. linear program. Therefore the objective becomes the find-
When more ingredients are used, some of the percent- ing of an acceptable ration formulation that would be as
ages become smaller because of the constraint that the close to the ideal solution as possible.

TABLE 3. Ingredient data1 for the multiple-objective (goal) programming variance reduction problem

Restrictions (%) Nutrient variances (%2)

Price
Ingredient Minimum Maximum (Dfl/cwt) Protein Lys Met
Corn 0.00 16.00 40.50 0.3600 0.0001 0.0001
Wheat meal 40.00 45.00 32.25 1.6900 0.0003 0.0000
Dutch peas 0.00 10.00 48.00 0.4900 0.0015 0.0001
Tapioca meal 0.00 25.00 27.80 0.1600 0.0000 0.0000
Rapeseed meal 0.00 10.00 32.80 0.3600 0.0051 0.0003
Fish meal 1.00 4.00 112.25 1.9600 0.0470 0.0108
Meat meal 0.00 5.00 61.80 2.5600 0.0240 0.3095
Feather meal 0.00 2.00 69.23 4.0000 0.0010 0.0001
Vegetable oil 0.00 3.00 54.50 0.0000 0.0000 0.0000
Sunflower seed meal 0.00 10.00 27.10 1.9600 0.0013 0.0008
Wheat middlings 0.00 10.00 22.00 0.6400 0.0008 0.0001
Heated soybean 0.00 10.00 55.70 1.2100 0.0061 0.0003
Soybean meal 0.00 30.00 49.10 1.6900 0.0041 0.0007
Corn gluten meal 0.00 5.00 28.00 13.6900 0.0028 0.0044
Alfalfa meal 0.00 5.00 23.30 0.3600 0.0005 0.0000
Animal fat 0.00 5.00 70.50 0.0000 0.0000 0.0000
Salt 0.00 1.00 11.00 0.0000 0.0000 0.0000
Limestone 0.00 3.00 4.40 0.0000 0.0000 0.0000
Dicalcium phosphate 0.00 3.00 8.80 0.0000 0.0000 0.0000
DL-Met 0.00 1.00 726.00 1.1720 0.0000 1.6858
Lys 0.00 1.00 440.00 1.9157 1.5600 0.0000
1
Data from an anonymous commercial Dutch manufacturer, 1997 (personal communication).
186 ZHANG AND ROUSH
FIGURE 1. The multiple-objective (goal) programming model for
feed formulation. Avail. = available.
The five rations in Table 4 were efficient solutions that
could be thought of as being near an ideal solution. How-
ever, four of the rations determined by minimizing vari-
ance(s) were not acceptable because of the ration cost.
Therefore, a compromise solution was made between the
first objective (least cost) and the second objective (mini-
mal variances).
Minimizing Protein, Methionine, and Lysine Vari-
ances. Minimizing the summation of the variances of pro-
tein, lysine, and methionine (w5 = w6 = w7 = 1) did not
result in a satisfactory ration. The cost of the ration was
43.13 Dfl/cwt. The results of treating ration cost as a hard
constraint and increasing by increments of 0.05 Dfl/cwt
are shown in Table 6 and Figure 2. An analysis was con-
ducted to minimize the three variances separately (through
the change of weights in the second objective).
Minimizing Protein Variance. The results of minimiz-
ing protein variance, with w5 = 1, w6 = 0, w7 = 0, are listed
in Table 6. Dutch peas, sunflower seed meal, and wheat
middlings were not included in the rations. Heated soy-
bean and alfalfa meals were included in the rations at high
ration costs. The differences were in the actual percentages
of the ingredients in these rations. The differences in nutri-
ent standard deviations of these rations were not apparent
when compared to similar cost rations. For example, the
protein standard deviation in the ration at a cost of 40.05
Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019
Dfl/cwt was 0.57%. The standard deviation of lysine in the
rations with cost of 39.60 Dfl/cwt was 0.01. The standard
deviation of methionine in the rations with cost of 39.90
Dfl/cwt was 0.02.
The CV plot is listed in Figure 3. The patterns indicate
that the variance of protein dominated the variances of
methionine. Therefore, minimizing the variance of protein
would not be considered a good choice because the stan-
dard deviation of methionine was near its highest value
in most of the rations.
Minimizing Lysine Variance. From the previous results,
it was suggested that by minimizing the variance of protein
alone in the feed formulation, the resulting rations might
not satisfy the animal requirements. Another analysis was
conducted by minimizing the variance of lysine. The re-
sults (with w5 = 0, w6 = 1, w7 = 0) are listed in Table 7.
Wheat middlings were not included in the rations, which
is similar to the results in Table 6. However, Dutch peas
and sunflower seed meal, as well as alfalfa meal and heated
soybean were included in some of the rations when the
tolerances of ration cost were increased.
The differences in nutrient standard deviation were rela-
tively small when compared to those in Table 6. For exam-
ple, the standard deviation of protein in the ration costing
40.045 Dfl/cwt was 0.5771 (Table 7) compared to 0.5695
(Table 6). The standard deviation of lysine in the rations
of cost of 39.595 Dfl/cwt was 0.0136 (Table 7) as compared
to 0.0138 (Table 6). The standard deviation of methionine
in the rations with a cost of 39.895 Dfl/cwt was 0.0100
(Table 7) as compared to 0.0167 (Table 6).
The CV plot is shown in Figure 4. In most rations, the
CV of protein were below 3%, (similar to Figure 3); the
CV of lysine were below 1.5% (similar to Figure 3). The CV
of methionine improved compared to those in Figure 3,
with all values falling below 2.5%. This result suggests
that minimizing the variance of lysine with a ration cost
trade-off was a better choice than minimizing protein vari-
ance or minimizing variances of protein, methionine,
and lysine.
Minimizing Methionine Variance. Another analysis
was conducted by minimizing the variance of methionine.
The results are listed in Table 8 (w5 = 0, w6 = 0, w7 = 1).
The ingredient percentages had a similar pattern in com-
parison to those in Tables 6 and 7. Dutch peas and wheat
middlings were not included in the rations. Heated soy-
 MULTIPLE-OBJECTIVE PROGRAMMING FOR FEED FORMULATION 187
TABLE 4. Efficient solutions for the multiple-objective (goal) programming variance reduction problem

Minimal nutrient variance1

Minimal ration Sum of
Ingredient cost Protein Lysine Methionine three

Ration cost, Dfl/cwt 39.34 43.65 42.14 42.98 43.13

Standard deviation
Protein, % 0.6156 0.5395 0.5528 0.5806 0.5399
Lys, % 0.0157 0.0142 0.0121 0.0134 0.0141
Met, % 0.0067 0.0161 0.0087 0.0043 0.0142
Calculated nutrient means
ME, kcal/kg 3,100.00 3,100.00 3,100.00 3,100.00 3,100.00
Protein, % 20.28 20.00 20.00 20.00 20.00
Lys, % 1.07 1.07 1.07 1.07 1.07
Met, % 0.48 0.52 0.48 0.48 0.52

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

Met + Cys, % 0.79 0.79 0.79 0.79 0.79
Thr, % 0.62 0.67 0.67 0.67 0.67
Trp, % 0.19 0.19 0.19 0.19 0.19
Ca, % 0.71 0.71 0.71 0.71 0.60
Total P, % 0.55 0.55 0.55 0.55 0.55
Available P, % 0.30 0.30 0.30 0.30 0.30
Na, % 0.14 0.16 0.16 0.16 0.14
Ingredient percentages
Corn 1.00 5.47 4.52 5.05 5.43
Wheat meal 40.00 40.00 40.00 42.49 40.00
Dutch peas 0.00 6.58 8.07 10.00 7.25
Tapioca meal 20.80 14.33 12.93 7.93 14.66
Rapeseed meal 3.34 4.36 1.88 1.26 4.36
Fish meal 1.00 4.00 1.27 1.31 4.00
Meat meal 0.00 2.71 1.34 0.03 2.36
Feather meal 0.17 0.67 1.46 1.47 0.69
Vegetable oil 3.00 0.09 0.64 0.02 3.00
Sunflower seed meal 0.00 0.43 3.10 0.26 0.16
Wheat middlings 0.00 1.92 0.00 1.41 1.94
Heated soybean 0.00 4.90 4.34 10.00 4.99
Soybean meal 19.88 5.33 9.56 6.66 5.50
Corn gluten meal 5.00 0.29 2.30 1.70 0.28
Alfalfa meal 0.00 2.18 1.11 2.77 2.22
Animal fat 3.08 5.00 5.00 5.00 1.70
Salt 0.30 0.21 0.31 0.34 0.17
Limestone 0.95 0.64 0.84 0.86 0.38
Dicalcium phosphate 0.95 0.32 0.75 0.86 0.36
DL-Met 0.17 0.24 0.20 0.22 0.24
Lys 0.36 0.34 0.39 0.38 0.33
1
Underline represents the minimal nutrient variance for the nutrient requested.

bean, sunflower seed meal, and alfalfa meal were included
in the rations when high ration cost was tolerated, but the
percentages in these rations were different.
The CV plot is listed in Figure 5, which is different from
those in Figure 3 or 4. In Figure 5, the CV of methionine
and lysine were both below 1.5%; however, the CV of
protein were, in general, around 3%. This finding suggests
that minimizing the variance of methionine was a better
choice in the process of trading-off with ration cost, because
CV of methionine and lysine decreased as the ration cost
increased. The minimization of the summation of variances
of protein, lysine, and methionine could be achieved by
assigning a higher weight for methionine than lysine. For
example, w5 = 1, w6 = 10, w7 = 100.
Comparing the efficient solutions listed in Tables 6, 7,
and 8, it is suggested that the ration listed in Table 8 with

TABLE 5. Example for nutrient variance reduction

Protein, % Ration, %
Ingredient Mean SD I II

Corn 8.60 0.60 70.00 63.00

Soybean 48.00 1.30 30.00 26.00
Peas 23.00 1.70 0 11.00
Total 100.00 100.00
Protein, %
Mean 20.42 20.42
SD 0.3285 0.2922
 188

TABLE 6. Nutrient values and ingredient percentages vs. ration cost by minimizing the variance of protein

Ration cost (Dfl/cwt)

Nutrient 39.35 39.40 39.45 39.50 39.55 39.60 39.65 39.70 39.75 39.80 39.85 39.90 39.95 40.00 40.05
Calculated nutrient means
ME, kcal/kg 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100
Protein, % 20.29 20.54 20.32 20.10 20.32 20.29 20.26 20.34 20.37 20.38 20.36 20.34 20.26 20.17 20.09
Lys, % 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07
Met, % 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48
Met + Cys, % 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79
Thr, % 0.62 0.63 0.64 0.66 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67
Trp, % 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19
Ca, % 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71
Available P, % 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30
Nutrient SD
Protein, % 0.6154 0.6081 0.6017 0.5958 0.5907 0.5870 0.5838 0.5811 0.5790 0.5773 0.5756 0.5740 0.5725 0.5710 0.5695
Lys, % 0.0157 0.0151 0.0146 0.0142 0.0140 0.0138 0.0137 0.0135 0.0133 0.0132 0.0132 0.0132 0.0132 0.0132 0.0131
Met, % 0.0066 0.0093 0.0094 0.0096 0.0148 0.0151 0.0153 0.0165 0.0166 0.0166 0.0167 0.0167 0.0167 0.0167 0.0167
Ingredient percentages
Corn 0.98 1.23 2.51 3.78 4.10 4.24 4.38 4.31 4.28 4.27 4.30 4.32 4.46 4.61 4.75
Wheat meal 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00
Tapioca meal 20.81 20.84 20.58 20.32 20.33 18.74 17.14 16.83 16.86 16.89 16.90 16.92 16.93 16.95 16.96
Rapeseed meal 3.35 3.88 4.16 4.43 4.94 5.22 5.50 5.57 5.48 5.43 5.44 5.45 5.46 5.46 5.46
Fish meal 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
ZHANG AND ROUSH

Meat meal 0.05 1.22 1.30 1.38 2.47 2.53 2.58 2.82 2.84 2.86 2.87 2.87 2.87 2.88 2.88
Feather meal 0.17 0.20 0.33 0.47 0.51 0.44 0.37 0.34 0.35 0.37 0.41 0.46 0.44 0.41 0.38
Vegetable oil 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00
Heated soybean 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.77 1.90 2.63 2.76 2.90 3.01 3.11 3.22
Soybean meal 19.85 18.38 17.05 15.71 14.25 13.49 12.73 11.92 11.20 10.74 10.67 10.60 10.51 10.42 10.33
Corn gluten meal 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 4.95 4.82 4.69 4.56 4.44 4.31
Alfalfa meal 0.00 0.00 0.00 0.00 0.00 1.56 3.14 3.44 3.34 3.28 3.26 3.24 3.22 3.20 3.18
Animal fat 3.08 2.78 2.59 2.39 2.10 2.57 3.04 2.94 2.69 2.53 2.51 2.50 2.48 2.46 2.45
Salt 0.31 0.28 0.27 0.27 0.25 0.24 0.24 0.23 0.23 0.23 0.23 0.23 0.23 0.23 0.23
Limestone 0.95 0.87 0.86 0.86 0.79 0.72 0.65 0.62 0.62 0.63 0.63 0.63 0.63 0.63 0.63
Dicalcium phosphate 0.94 0.81 0.80 0.80 0.67 0.65 0.63 0.60 0.60 0.60 0.60 0.60 0.60 0.60 0.60
DL-Met 0.17 0.16 0.17 0.17 0.17 0.17 0.19 0.17 0.17 0.17 0.17 0.17 0.18 0.14 0.18
Lys 0.36 0.36 0.39 0.42 0.42 0.43 0.44 0.44 0.44 0.44 0.44 0.44 0.44 0.44 0.44

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

 TABLE 7. Nutrient values and ingredient percentages vs. ration cost by minimizing the variance of lysine

Ration cost, Dfl/cwt

Nutrient 39.35 39.40 39.45 39.50 39.55 39.60 39.65 39.70 39.75 39.80 39.85 39.90 39.95 40.00 40.05

Calculated nutrient means

ME, kcal/kg 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100
Protein, % 20.29 20.42 20.19 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.05
Lys, % 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07
Met, % 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48
Met + Cys, % 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79
Thr, % 0.62 0.63 0.64 0.66 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67 0.67
Trp, % 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19
Ca, % 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71
Available P, % 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30
Nutrient SD
Protein, % 0.6154 0.6081 0.6017 0.5958 0.5914 0.5887 0.5865 0.5841 0.5814 0.5792 0.5782 0.5779 0.5776 0.5775 0.5771
Lys, % 0.0157 0.0151 0.0146 0.0142 0.0139 0.0136 0.0133 0.0131 0.0130 0.0128 0.0127 0.0126 0.0126 0.0125 0.0125
Met, % 0.0066 0.0082 0.0082 0.0086 0.0098 0.0098 0.0098 0.0101 0.0105 0.0108 0.0105 0.0100 0.0095 0.0090 0.0093
Ingredient percentages
Corn 0.98 1.49 2.81 4.02 4.47 4.53 4.60 4.68 4.77 4.85 4.88 4.86 4.85 4.83 4.75
Wheat meal 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00
Dutch peas 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.45 0.93 1.45 1.92
Tapioca meal 20.81 20.77 20.49 20.25 20.15 19.98 19.77 19.08 18.07 17.24 17.37 17.09 16.78 16.45 16.21
Rapeseed meal 3.35 3.82 4.08 4.38 4.56 4.11 3.55 3.26 3.16 3.00 2.55 2.43 2.37 2.28 2.26
Fish meal 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
Meat meal 0.05 0.94 0.98 1.12 1.43 1.45 1.47 1.53 1.63 1.70 1.66 1.56 1.45 1.34 1.41
Feather meal 0.17 0.22 0.37 0.50 0.55 0.59 0.65 0.67 0.65 0.65 0.70 0.70 0.69 0.68 0.67
Vegetable oil 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00
Sunflower seed 0.00 0.00 0.00 0.00 0.00 0.52 1.19 1.67 2.02 2.38 2.79 2.82 2.78 2.77 2.76
Heated soybean 0.00 0.00 0.00 0.00 0.48 1.20 1.82 2.19 2.38 2.71 3.74 3.86 3.92 3.89 3.91
Soybean meal 19.85 18.41 17.09 15.74 14.65 14.01 13.40 12.74 12.03 11.33 10.71 10.64 10.61 10.64 10.53
Corn gluten meal 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00
Alfalfa meal 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.55 1.46 2.17 1.88 1.85 1.86 1.85 1.82
Animal fat 3.08 2.81 2.62 2.41 2.18 2.08 2.00 2.12 2.37 2.54 2.29 2.28 2.28 2.31 2.29
MULTIPLE-OBJECTIVE PROGRAMMING FOR FEED FORMULATION

Salt 0.30 0.28 0.28 0.28 0.27 0.27 0.27 0.27 0.26 0.26 0.26 0.26 0.27 0.27 0.27
Limestone 0.94 0.89 0.88 0.87 0.85 0.86 0.86 0.83 0.79 0.76 0.78 0.79 0.79 0.80 0.80
Dicalcium phosphate 0.94 0.84 0.84 0.83 0.79 0.80 0.79 0.78 0.76 0.75 0.76 0.77 0.78 0.79 0.78
DL-Met 0.17 0.17 0.17 0.17 0.17 0.17 0.19 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17
Lys 0.36 0.37 0.40 0.43 0.44 0.45 0.45 0.46 0.47 0.48 0.48 0.47 0.47 0.47 0.46
189

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

 190

TABLE 8. Nutrient means and ingredient percentages vs. ration cost by minimizing the variance of methionine

Ration cost (Dfl/cwt)

Name 39.35 39.40 39.45 39.50 39.55 39.60 39.65 39.70 39.75 39.80 39.85 39.90 39.95 40.00 40.05
Calculated nutrient means
ME, kcal/kg 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100 3,100
Protein, % 20.28 20.03 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00 20.00
Lys, % 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07 1.07
Met, % 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48 0.48
Met + Cys, % 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.79 0.80 0.80 0.80 0.81 0.81 0.81
Thr, % 0.62 0.63 0.64 0.64 0.64 0.64 0.64 0.64 0.65 0.66 0.67 0.67 0.67 0.67 0.67
Trp, % 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19 0.19
Ca, % 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.60 0.60 0.60 0.60
Available P, % 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30
Nutrient SD
Protein, % 0.6155 0.6085 0.6042 0.6008 0.5978 0.5978 0.5921 0.5885 0.5862 0.5841 0.5821 0.6030 0.6365 0.6358 0.6346
Lys, % 0.0157 0.0152 0.0148 0.0145 0.0142 0.0142 0.0138 0.0136 0.0135 0.0133 0.0136 0.0132 0.0133 0.0133 0.0132
Met, % 0.0067 0.0064 0.0062 0.0061 0.0060 0.0060 0.0057 0.0056 0.0055 0.0055 0.0054 0.0053 0.0052 0.0052 0.0051
Ingredient percentages
Corn 1.02 2.36 2.63 2.71 2.78 2.78 2.86 2.90 3.34 3.78 4.22 4.48 4.58 4.60 4.58
Wheat meal 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 40.00 41.98 45.00 45.00 45.00
Tapioca meal 20.80 20.52 20.42 20.34 20.27 20.27 19.31 16.56 16.41 16.26 16.12 14.95 12.51 12.58 12.59
Rapeseed meal 3.35 3.61 3.61 3.55 3.48 3.48 3.50 4.04 4.19 4.34 4.50 4.19 3.58 3.33 3.35
Fish meal 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
ZHANG AND ROUSH

Meat meal 0.01 0.01 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.03 0.03 0.03
Feather meal 0.17 0.32 0.35 0.36 0.37 0.37 0.33 0.28 0.48 0.68 0.89 0.99 1.05 1.07 1.13
Vegetable oil 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 3.00 2.97 2.86 2.85
Sunflower seed 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.11 0.34 0.35
Heated soybean 0.00 0.00 0.83 1.85 2.86 2.86 4.44 3.44 3.45 3.46 3.47 3.57 3.52 4.09 4.12
Soybean meal 19.85 18.53 17.67 16.92 16.17 16.17 14.73 14.25 13.66 13.06 12.46 11.90 11.49 11.13 11.14
Corn gluten meal 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 4.87
Alfalfa meal 0.00 0.00 0.00 0.00 0.00 0.00 0.85 3.62 3.62 3.63 3.63 3.79 4.08 3.91 3.90
Animal fat 3.08 2.89 2.68 2.47 2.26 2.26 2.20 3.27 3.19 3.12 3.04 2.76 2.72 2.70 2.72
Salt 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.30 0.29 0.29 0.29 0.29 0.29 0.29 0.29
Limestone 0.95 0.95 0.95 0.95 0.95 0.95 0.91 0.79 0.79 0.79 0.79 0.51 0.50 0.51 0.51
Dicalcium phosphate 0.95 0.95 0.95 0.95 0.95 0.95 0.94 0.91 0.92 0.92 0.92 0.91 0.91 0.91 0.91
DL-Met 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.17 0.18 0.17 0.18 0.18 0.18
Lys 0.36 0.40 0.40 0.40 0.40 0.40 0.41 0.42 0.43 0.45 0.46 0.47 0.48 0.48 0.48

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

 MULTIPLE-OBJECTIVE PROGRAMMING FOR FEED FORMULATION 191

Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019

FIGURE 2. Nutrient coefficient of variation vs. ration cost by minimizing the sum of variances of protein, lysine, and methionine.

FIGURE 3. Nutrient coefficient of variation vs. ration cost by minimizing the variance of protein.

FIGURE 4. Nutrient coefficient of variation vs. ration cost by miniziming the variance of lysine.
192 ZHANG AND ROUSH
FIGURE 5. Nutrient coefficient of variation vs. ration cost by minimizing the variance of methionine.
a ration cost of 39.495 Dfl/cwt was a best compromise
solution. The ration cost was between 39.00 (goal) and
40.05 Dfl/cwt (upper limit), and the variances of protein,
lysine, and methionine were 111, 120, and 141% of their
own ideal variance, respectively. Of course, any particular
decision-maker can choose according to preference among
the rations in Tables 6, 7, and 8.
In summary, the MOP model can handle several con-
flicting objectives simultaneously (e.g., feasibility, nutrient
ratios, least cost, and minimum variances) as compared to
conventional linear programming that can solve only for
single objectives such as least-cost rations. The acceptable
solution was accomplished through the process of solving
a series of linear and nonlinear programs and by interac-
tion with the decision-maker(s). The MOP model allows
for flexible alternatives to infeasible rations in contrast to
conventional least-cost formulation.
ACKNOWLEDGMENTS
This research was funded by the Pennsylvania Depart-
ment of Agriculture with a grant entitled “Strategic Feed
Formulation for Nutrient Management (ME# 449311).” The
authors gratefully acknowledge the assistance of T. L. Cra-
vener in the development of the manuscript.
REFERENCES
Emmert, J. L., and D. H. Baker, 1997. Use of the ideal protein
concept for precision formulation of amino acid levels in
broiler diets. J. Appl. Poult. Res. 6:462–270.
Downloaded from https://academic.oup.com/ps/article-abstract/81/2/182/1576411 by guest on 04 June 2019
Ignizio, J. P., and T. M. Cavalier, 1994. Linear Programming.
Prentice Hall, Englewood Cliffs, NJ.
Lara, P., 1993. Multiple objective fractional programming and
livestock ration formulation: A cast study for dairy cow diets
in Spain. Agric. Syst. 41:321–334.
Lara, P., 1996. Linking production theory and multiobjective frac-
tional programming as a support tool for animal diet formula-
tion. Advances in multiple objective and goal programming.
Pages 301–309 in: Proceedings of the Second International
Conference on Multi-Objective Programming and Goal Pro-
gramming. R. Calallero, F. Ruiz, and R. E. Steuter, ed. Torrem-
olinos, Spain. Springer-Verlag, New York.
Lara, P., and C. Romero, 1992. An interactive multigoal program-
ming model for determining livestock rations: An application
to dairy cows in Andalusia, Spain. J. Opl. Res. Soc. 43:945–953.
Lara, P., and C. Romero, 1994. Relaxation of nutrient require-
ments on livestock rations through interactive multigoal pro-
gramming. Agric. Syst. 45:443–453.
McCoy, R. A., K. C. Behnke, J. D. Hancock, and R. R. McEllhiney,
1994. Effect of mixing uniformity on broiler chick performance.
Poultry Sci. 73:443–451.
Mollaghasemi, M., and J. Pet-Edwards, 1997. Technical Briefing:
Making Multiple-Objective Decisions. IEEE Computer Society
Press, Los Alamito, CA.
Oberstone, J., 1997. Management Science: Concepts, Insights and
Applications. West Publishing Co., New York, NY.
Satoris, W., and W. Spruill, 1974. Goal programming and working
capital management. Financial Mgmt. 3:67–74.
Tozer, P. R., and J. R. Stokes, 2001. A multi-objective program-
ming approach to feed ration balancing and nutrient manage-
ment. Agric. Syst. 67:201–215.
Steurer, R., 1984. Sausage blending using multiple objective pro-
gramming. Management Sci. 30:1376–1384.
Winston, W. L., 1994. Operations Research, Applications and
Algorithms. Third edition. Duxbury Press, Belmont, CA.
Zander, P., and H. Kächele, 1999. Modelling multiple objectives of
land use for sustainable development. Agric. Syst. 59:311–325.