Research Paper

Academic Platform Journal of Engineering and Smart Systems (APJESS) 11(1), 35-40, 2023
ISSN: 2822-2385
Received: 15-Nov-2022 Accepted: 20-Dec-2022
https://doi.org/10.21541/apjess.1205309

Implementation of Fuzzy Linear Programming Approach for More Accurate

Demand Forecasting in a Make-to-Stock Company
*1
Çağatay TEKE

1
Department of Industrial Engineering, Bayburt University, Bayburt, Türkiye, [email protected]

Abstract

Demand forecasting is of crucial importance for make-to-stock companies because product demand is uncertain and it changes
with time. Fuzzy linear programming (FLP) can be an optimum approach for such uncertain situations. In this study, the FLP
model was used for more accurate demand forecasting in a make-to-stock company. Demand forecasting study was carried out
according to the FLP method and linear programming (LP) method. Solutions of FLP and LP were compared in terms of imputed
shortage cost, inventory carrying cost, and net profit. Results show that the applied FLP method is more advantageous than LP
as it provides a 67% decrease in costs and a 15% increase in net profit.

Keywords: Demand Forecasting; Fuzzy Linear Programming; Make-to-Stock.

1. INTRODUCTION Thanakijkasem [4] developed a model with hierarchical

genetic algorithm to optimize the steel production. Figueroa-
Determination of product demand is very important in terms Garcia et al. [5] used fuzzy linear programming (FLP) and
of production planning since companies make the production interval fuzzy set approach for mixed production planning
plan according to product demand. On the other hand, problem. Torabi et al. [6] used fuzzy hierarchical production
demand amounts in make-to-stock companies can’t be planning approach for solving multi-level production
known precisely. Therefore, they make the production plan planning problem. Bilgen [7] utilized fuzzy mathematical
in accordance with the estimated demand value. programming approach for production and distribution
planning problem. Ertuğrul and Tuş [8] investigated the
Make-to-stock companies aim to forecast the product interactive FLP approach by utilizing Zimmermann,
demand with a minimum error as more realistic demand Werners, Chanas and Verdegay approaches. Their
forecasting enables to decrease in the inventory carrying cost application was performed in the textile industry.
and imputed shortage cost. In addition, when companies
achieve this aim, costs based on the deterioration of product In recent years, FLP has been used in areas such as finance,
inventory decline. However, improving demand forecast energy, and transportation [9-15]. In the literature, no study
accuracy is a difficult issue because product demand is has been found in which FLP has been compared with
unknown and it varies from period to period. Thus, fuzzy classical LP in the demand forecasting problem of a make-
logic is a suitable approach to deal with decision problems to-stock company. In this respect, it is expected that this
involving such uncertainties and linear programming (LP) study will make a significant contribution to the literature.
[1]. Zimmermann introduced a fuzzy approach regarding LP
problems in 1974. In addition, Lai and Hwang, Tong In this paper, production planning problem in a make-to-
Shaocheng, Buckley, among others, cope with the fact that stock company was investigated and a demand forecasting
all parameters are fuzzy [2]. study was conducted. FLP approach was used to forecast the
product demand accurately. Because of the fact that both
When evaluated similar studies, Azizi et al. [3] developed objective function and constraints include uncertainty,
adaptive neuro fuzzy inference system to predict the Zimmerman’s approach was used. Estimated product
throughput of production systems. It was used because of demands were obtained from the models of classical LP and
factors include uncertainty such as set up time and demand. FLP. Production planning was performed for each model by
It was applied in tile industry. Obtained results revealed that using these demand values for three periods. Results were
adaptive neuro fuzzy inference system is appropriate for compared with the aspect of inventory carrying cost, imputed
prediction of throughput in uncertainty. Worapradya and shortage cost, and net profit.

Attribution (CC BY) license https://creativecommons.org/licenses/by/4.0/ https://dergipark.org.tr/tr/pub/apjess

 Çağatay TEKE
Implementation of Fuzzy Linear Programming Approach for More Accurate Demand Forecasting in a Make-to-Stock Company

2. FUZZY SETS 3. FUZZY LINEAR PROGRAMMING

In classical sets, there are two states, 0 and 1. In other words, 3.1. Linear Programming (LP)
an element either belongs to a set or doesn’t [16]. For
instance, it is supposed that there is a car. The headlights, air LP is a type of mathematical programming. The aim of
conditioner and windscreen wiper of the car are out of mathematical programming is to determine the extremum
service. However, the engine of the car is running and it is point of a function under some constraints. Both objective
moving. In that case, it can’t be reached a conclusion that function and constraints are linear in LP [19]. LP model can
either the car is out of order exactly or it runs precisely. be loosely formulated as follows:
Therefore, it is more suitable to characterize the car as a
slight level failed and middle level failed. As is also Max CT x
understood from this example, an element can belong to a set subject to (1)
with any value between 0 and 1. This value is entitled as the
Ax  b
degree of membership and symbolized with “µ” for fuzzy
sets fundamental to fuzzy logic. Also, degrees of x0
membership are expressed with membership functions.
in which x represents the vector of variables, C is the vector
It can be propounded that there are three basic reasons for of coefficients in objective function, (.) T is the matrix
using fuzzy logic. The first of these is subjectivity. transpose, A is a matrix of coefficients in constraints, b is a
Judgements related to a situation can vary from person to vector of right-hand sides of the constraints. The expression
person. For example, we assume there are three age sets; to be max or min is called objective function.
young, middle and old. While thirty age belongs to ‘young’
set for a person, it belongs to ‘middle’ set for another person. LP is of crucial importance and it is used in many fields such
Secondly, whole elements of a set can’t belong to it with the as engineering and economy [20]. LP problems can be
same degree of membership. In addition, an element can solved with LINDO optimization software. If the objective
belong to sets more than one [17]. Thirdly, real world function, right-hand sides or coefficients of constraints are
problems include many variables. Each variable’s effect to uncertain, LP isn't appropriate to solve the problem.
problem and interactions among them can’t be determined
exactly. In other words, this is very difficult to model this Many types of FLP were formed. Some methods were
type of problems. Thus, fuzzy logic can be an ideal approach developed for converting them to LP. Finally, they were
for solving them. solved with optimization software [21].

For instance, it is presumed that there are five sets about 3.2. Fuzzy Linear Programming (FLP)
temperature; very low, low, middle, high and very high. In
this case, we have five fuzzy sets and five membership Sometimes, coefficient of the variable belongs to a problem
functions. These are illustrated below: depends on other parameters. Also, it can’t be evaluated
exactly and can determine subjectively. Thus, coefficients in
problem can include uncertainty. In this case, classical LP
falls behind with the solution of the problem. FLP can be an
ideal approach for such problems [22]. Formulation of the
FLP model can be considered as follows:

~ ~
Max Z = CT x
subject to
(2)
Figure 1. Fuzzy sets and membership functions ~ ~~
Axb
As shown in Figure 1, each element of a set can’t belong to x0
it with the same degree of membership. For example,
whereas 15Cº belongs to ‘middle’ set with 1 degree of in which tilde (~) is fuzzy symbol differently from LP model.
membership, 8Cº belongs to it with 0.4 degree. Also, 22Cº To solve this problem, it is necessary to find the probability
belongs to ‘high’ set with 0.2 degree of membership and distribution of the optional objective function Z. To deal with
‘middle’ set with 0.6 degree. In other words, it belongs to this problem, fuzzy constraints and fuzzy objective function
two sets. convert into crisp ones [23].

In fuzzy logic, triangular and trapezoidal membership 3.3. Zimmermann Method

functions are used generally. Additionally, membership
functions such as gaussian and sigmoid are also available. This method is one of the FLP models. Both objective
While determining to suitable membership function for a function and constraints are fuzzy in this method. An FLP
problem, capability of representation the data in it of model developed by Zimmermann is indicated as follows
membership function must be taken into account [18]. [21]:

Academic Platform Journal of Engineering and Smart Systems 11(1), 35-40, 2023 36
 Çağatay TEKE
Implementation of Fuzzy Linear Programming Approach for More Accurate Demand Forecasting in a Make-to-Stock Company

Then, the problem can be stated as LP as follows:

~
C T x  b0 Max λ
~
( Ax )i  bi
s.t.
μ 0 (x )  λ
(3) (9)
x  0 i = 1, 2 , ... , m μ i (x )  λ
λ  0,1
A symmetrical model in which the objective function
becomes a constraint is the inequality. To arrive at a general
formulation, this inequality is converted into matrix form as This problem is illustrated by the membership functions of
follows: the fuzzy objective function and fuzzy constraints as follows:
Max λ
~ s.t.
− C T x − b0 (4)
b0 − Cx
1− λ
in which d0
(10)
( Ax )i − bi
1−  λ , i
− C   − b0  di
B=  b=  (5)
 Ai   bi  λ  0 ,1
x0
The constraint's inequalities mean "as small as or equal to"
can be allowed to violate right-hand side b by expanding a The final form of the FLP model is obtained after some
value. The degree of violation is represented by membership simplifications as follows:
function as:
Max λ
 0 ,i f Cx  b0 − d 0  s.t.
 
 b0 − Cx 
μ 0 (x ) = 1 − ,if b 0 − d 0  Cx  b0  (6) C T x − λd 0  b0 − d 0 (11)
 d0  ( Ax )i + λd i  bi + d i , i
 ,i f Cx  b0  λ  0 ,1
 1 
x0
 0 , if ( Axi )  bi + di 
 
 ( Ax )i − bi
4. IMPLEMENTATION

μi (x ) = 1 − ,if bi  ( Ax )i  bi + di  (7)
 di  Demand forecasting study was conducted with data obtained

 1 , if ( Ax )i  bi 
 from a make-to-stock company. The problem was built as
FLP model because demand of the product types and
in which d is a matrix of admissible violation. This problem expected profit are uncertain. Solution of this model gave
can be expressed as follows using an auxiliary variable estimated demand values for each product type per month to
represented by λ: maximize the total profit. Since the company follows make-
to-stock strategy, estimated demand values are equal to
μ 0 (x )  λ
production amounts. Production data and constraints are
given in Table 1.
μ i (x )  λ (8)
λ  0 ,1 4.1. Classical LP Model and Solution

Problem was modelled as monthly basis by using related data

in Table 1. The classical LP model of the problem is given
Eq. (12).
Table 1. Application data
Variables for products X1 X2 X3 X4 X5 X6 X7 X8 X9
Profits (TRY per tonne) 1509 1509 503 1509 1006 503 1509 1006 503
Expected demands (tonne per month) 25 100 150 30 150 200 30 250 80
Tolerances for demands (tonne per month) 5 50 40 5 40 50 5 40 20
Labor usage (hour per tonne) 0.333 0.333 0.333 0.283 0.283 0.283 0.3 0.3 0.3
Expected profit (TRY) 1,200,000
Tolerance for profit (TRY) 200,000
Monthly production capacity (tonne) 1150
Monthly labor capacity (hour) 364

Academic Platform Journal of Engineering and Smart Systems 11(1), 35-40, 2023 37
 Çağatay TEKE
Implementation of Fuzzy Linear Programming Approach for More Accurate Demand Forecasting in a Make-to-Stock Company

𝑀𝑎𝑥 1509𝑥1 + 1509𝑥2 + 503𝑥3 + 1509𝑥4 + 1006𝑥5 + 503𝑥6 + 1509𝑥7 + 1006𝑥8 + 503𝑥9
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
𝑥1 ≥ 25
𝑥2 ≥ 100
𝑥3 ≥ 150
𝑥4 ≥ 30
𝑥5 ≥ 150
𝑥6 ≥ 200 (12)
𝑥7 ≥ 30
𝑥8 ≥ 250
𝑥9 ≥ 80
0.333𝑥1 + 0.333𝑥2 + 0.333𝑥3 + 0.283𝑥4 + 0.283𝑥5 + 0.283𝑥6 + 0.3𝑥7 + 0.3𝑥8 + 0.3𝑥9 ≤ 364
𝑥1 + 𝑥2 + 𝑥3 + 𝑥4 + 𝑥5 + 𝑥6 + 𝑥7 + 𝑥8 + 𝑥9 ≤ 1150
𝑥𝑖 ≥ 0
𝑖 = 1,2, . . . ,9

𝑀𝑎𝑥 𝜆
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜
1509𝑥1 + 1509𝑥2 + 503𝑥3 + 1509𝑥4 + 1006𝑥5 + 503𝑥6 + 1509𝑥7 + 1006𝑥8 + 503𝑥9 − 200,000𝜆 ≥ 1,000,000
𝑥1 + 5𝜆 ≤ 30
𝑥2 + 50𝜆 ≤ 150
𝑥3 + 40𝜆 ≤ 190
𝑥4 + 5𝜆 ≤ 35
𝑥5 + 40𝜆 ≤ 190
𝑥6 + 50𝜆 ≤ 250
𝑥7 + 5𝜆 ≤ 35 (13)
𝑥8 + 40𝜆 ≤ 290
𝑥9 + 20𝜆 ≤ 100
0.333𝑥1 + 0.333𝑥2 + 0.333𝑥3 + 0.283𝑥4 + 0.283𝑥5 + 0.283𝑥6 + 0.3𝑥7 + 0.3𝑥8 + 0.3𝑥9 ≤ 364
𝑥1 + 𝑥2 + 𝑥3 + 𝑥4 + 𝑥5 + 𝑥6 + 𝑥7 + 𝑥8 + 𝑥9 ≤ 1150
𝜆 ∈ [0,1]
𝑥𝑖 ≥ 0 𝑖 = 1,2, . . . ,9

LINDO optimization software was used to solve the LP LINDO software was used to solve the FLP model of the
model of the problem. The relevant solution is given in Table problem. The relevant solution is given in Table 3.
2.
Table 3. Results of FLP model
Table 2. Results of LP model
Variable Value Reduced Cost
Variable Value Reduced Cost λ 0.234 0
Objective Function 1,101,570 TRY - X1 28.832 0
X1 160 0 X2 138.321 0
X2 100 0 X3 180.657 0
X3 150 0 X4 33.832 0
X4 30 0 X5 180.657 0
X5 150 0 X6 177.885 0
X6 200 0 X7 33.832 0
X7 30 0 X8 280.657 0
X8 250 0 X9 95.329 0
X9 80 0
4.3. Comparison Between FLP and LP
4.2. FLP Model and Solution
Quarterly production plans of each model were made by
Problem was modelled as monthly basis by using related data using estimated demand values obtained from LP and FLP.
in Table 1. The FLP model of the problem is given Eq. (13).

Academic Platform Journal of Engineering and Smart Systems 11(1), 35-40, 2023 38
 Çağatay TEKE
Implementation of Fuzzy Linear Programming Approach for More Accurate Demand Forecasting in a Make-to-Stock Company

Outputs of production plans were compared in terms of If it meets back orders next month, imputed shortage cost for
inventory carrying cost, imputed shortage cost and net profit. stock-out products is equal to 50% of unit profit. If it meets
after two months, the cost is equal to 75% of unit profit.
The company can carry inventory maximum 3 months Otherwise, the cost is equal to unit profit.
because of the deterioration problem of products. Monthly
inventory carrying cost for the company is 12% of product Production plans of LP and FLP are shown in Table 4 and
profit per tonne. The company practices back order strategy. Table 5.

Table 4. Production planning of LP model

P D IS P D IS P D IS
Descriptions
(t-2) (t-2) (t-2) (t-1) (t-1) (t-1) (t) (t) (t)
X1 160 30 130 (I) 160 27 263 (I) 160 29 394 (I) t: Time period (month), P: Production (tonne),
D: Demand (tonne), IS: Inventory Status, I: Inventory, S:
X2 100 120 200 (S) 100 113 201+130 (S) 100 138 202+131+380 (S)
Stock-out
200: 20 items can’t be met in this period, they can be met next
X3 150 147 3 (I) 150 149 4 (I) 150 162 80 (S)
period.
X4 30 28 2 (I) 30 33 10 (S) 30 31 -
202+131+380: 20 items are stock-out since two period, 13
X5 150 160 100 (S) 150 152 101+20 (S) 150 157 102+21+70 (S) items are stock-out since one period and 38 items are stock-
out in last period.
X6 200 190 10 (I) 200 195 15 (I) 200 184 31 (I) Profit Status

X7 30 31 10 (S) 30 32 11+20 (S) 30 34 12+21+40 (S) Total Profit = ∑9x=1 Estimated production amount x Unit profit
Net Profit = Total Profit – (Inventory carrying cost + Imputed
X8 250 257 70 (S) 250 261 71+110 (S) 250 260 72+111+100 (S)
shortage cost)
Net Profit = 1,101,570 - (146,673.84 + 123,863.75) =
X9 80 83 30 (S) 80 84 31+40 (S) 80 89 32+41+90 (S)
831,032.4 TRY

Table 5. Production planning of FLP model

P D IS P D IS P D IS
Net Profit
(t-2) (t-2) (t-2) (t-1) (t-1) (t-1) (t) (t) (t)
X1 28.832 30 1.1680 (S) 28.832 27 0.664 (I) 28.832 29 0.496 (I)
X2 138.321 120 18.321 (I) 138.321 113 43.642 (I) 138.321 138 43.963 (I)
X3 180.657 147 33.657 (I) 180.657 149 65.314 (I) 180.657 162 83.971 (I)
X4 33.832 28 5.832 (I) 33.832 33 6.664 (I) 33.832 31 9.496 (I)
X5 180.657 160 20.657 (I) 180.657 152 49.314 (I) 180.657 157 72.971 (I) 958,115.4 TRY
12.1151 12.1152
X6 177.885 190 12.1150 (S) 177.885 195 177.885 184
+17.1150 (S) +17.1151+6.1150 (S)
X7 33.832 31 2.832 (I) 33.832 32 4.664 (I) 33.832 34 4.496 (I)
X8 280.657 257 23.657 (I) 280.657 261 43.314 (I) 280.657 260 63.971 (I)
X9 95.329 83 12.329 (I) 95.329 84 23.658 (I) 95.329 89 29.987 (I)

5. CONCLUSION on solutions of FLP and LP models were compared in terms

of costs and net profit. Results showed that when fuzzy logic
In this paper, demand forecasting problem of a make-to- and LP was used together, product demands were forecasted
stock company was addressed. Because the company follows more accurately. Therefore, inventory carrying cost and
make-to-stock strategy, it needs estimated product demand imputed shortage cost reduced by 67% totally and also net
values to develop a production planning. However, product profit increased by 15%.
demand cannot be known precisely and it varies from period
to period. Because of this uncertainty, fuzzy logic can be an Author contributions: Concept – Ç.T.; Data Collection
ideal approach for estimating the product demand with a &/or Processing – Ç.T.; Literature Search – Ç.T.; Writing –
minimum error. Ç.T.
Conflict of Interest: No conflict of interest was declared by
As estimation of product demand was more accurate, the author.
Zimmerman’s FLP approach was used. Amount of Financial Disclosure: The author declared that this study
production for each product type was obtained from the has received no financial support.
solutions of FLP and LP. Output of production plans based

Academic Platform Journal of Engineering and Smart Systems 11(1), 35-40, 2023 39
 Çağatay TEKE
Implementation of Fuzzy Linear Programming Approach for More Accurate Demand Forecasting in a Make-to-Stock Company

REFERENCES [13] I. U. Khan, and F. Rafique, "Minimum-cost capacitated

fuzzy network, fuzzy linear programming formulation,
[1] S. M. Guu and Y. K. Wu, "Two-phase approach for and perspective data analytics to minimize the
solving the fuzzy linear programming problems", operations cost of American airlines", Soft Computing,
Fuzzy Sets and Systems, vol. 107, pp. 191-195, 1999. vol. 25, pp. 1411–1429, 2021.
[2] A. N. Gani, C. Duraisamy, and C. Veeramani, "A note [14] E. Escobar-Gómez, J. L. Camas-Anzueto, S.
on fuzzy linear programming problem using L-R fuzzy Velázquez-Trujillo, H. Hernández-de-León, R.
number", International Journal of Algorithms, Grajales-Coutiño, E. Chandomí-Castellanos, and H.
Computing and Mathematics, vol. 3, pp. 93-106, 2009. Guerra-Crespo, "A linear programming model with
fuzzy arc for route optimization in the urban road
[3] A. Azizi, A. Y. Ali, and L. W. Ping, "Modelling
network", Sustainability, vol. 11, no. 23, pp. 6665,
production uncertainties using the adaptive neuro-
2019.
fuzzy inference system", South African Journal of
Industrial Engineering, vol. 26, no. 1, pp. 224-234, [15] A. Güler and H. Bircan, "Fuzzy linear programming
2015. and an application in BIMS production facility",
Turkish Studies - Economy, vol. 16, no. 2, pp. 823-837,
[4] K. Worapradya and P. Thanakijkasem, "Optimising
2021.
steel production schedules via a hierarchical genetic
algorithm", South African Journal of Industrial [16] G. Chen and T. T. Pham, Introduction to Fuzzy Sets,
Engineering, vol. 25, no. 2, pp. 209-221, 2014. Fuzzy Logic and Fuzzy Control Systems. Boca Raton:
CRC Press, 2001.
[5] J. C. Figueroa-García, D. Kalenatic, and C. A. Lopez-
Bello, "Multi-period mixed production planning with [17] B. Yıldız, Artificial Intelligence in Financial Analysis,
uncertain demands: fuzzy and interval fuzzy sets Detay Publishing, 2009.
approach", Fuzzy Sets and Systems, vol. 206, pp. 21-
[18] M. Mutingi and C. Mbohwa, "A fuzzy-based particle
38, 2012.
swarm optimisation approach for task assignment in
[6] S. A. Torabi, M. Ebadian, and R. Tanha, "Fuzzy home healthcare", South African Journal of Industrial
hierarchical production planning (with a case study)", Engineering, vol. 25, no. 3, pp. 84-95, 2014.
Fuzzy Sets and Systems, vol. 161, no. 11, pp. 1511-
[19] E. Dervişoğlu, "Fuzzy linear programming: review and
1529, 2010.
implementation", Master thesis, Sabanci University
[7] B. Bilgen, "Application of fuzzy mathematical Institute of Engineering and Science, Istanbul, 2005.
programming approach to the production allocation and
[20] J. L. Verdegay, "Progress on fuzzy mathematical
distribution supply chain network problem", Expert
programming: A personal perspective", Fuzzy Sets and
Systems with Applications, vol. 37, no. 6, pp. 4488-
Systems, vol. 281, pp. 219-226, 2015.
4495, 2010.
[21] C. S. Lee and C. G. Wen, "River assimilative capacity
[8] I. Ertuğrul and A. Tuş, "Interactive fuzzy linear
analysis via fuzzy linear programming", Fuzzy Sets and
programming and an application sample at a textile
Systems, vol. 79, pp. 191-201, 1996.
firm", Fuzzy Optimization and Decision Making, vol.
6, no. 1, pp. 29-49, 2007. [22] J. Ren and T. B. Sheridan, Optimization with Fuzzy
Linear Programming and Fuzzy Knowledge Base,
[9] S. Akbaş, T. E. Dalkilic, and T. G. Aksoy, "A study on
IEEE 3rd International Fuzzy Systems Conference, pp.
portfolio selection based on fuzzy linear
1389-1393, Orlando, 1994.
programming", International Journal of Uncertainty,
Fuzziness and Knowledge-Based Systems, vol. 30, no. [23] R. R. Gasimov and K. Yenilmez, "Solving fuzzy linear
2, pp. 211-230, 2022. programming problems with linear membership
functions", Turkish Journal of Mathematics, vol. 26,
[10] D. Yin, "Application of interval valued fuzzy linear
no. 4, pp. 375-396, 2002.
programming for stock portfolio optimization",
Applied Mathematics, vol. 9, pp.101-113, 2018.
[11] Ş. Emeç and G. Akkaya, "Developing a new
optimization energy model using fuzzy linear
programming", Journal of Intelligent and Fuzzy
Systems, vol. 40, pp. 9529–9542, 2021.
[12] P. Sefeedpari, S. Rafiee, M. Sharifi, and A. Akram,
"Detecting the best optimal dairy cattle herd sizes based
on energy consumption using fuzzy linear
programming model", Iranian Journal of Biosystem
Engineering, vol. 49, no. 1, pp. 149-159, 2018.

Academic Platform Journal of Engineering and Smart Systems 11(1), 35-40, 2023 40