2019 - Artículo - Hybrid Model

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

International Journal of Industrial Engineering, 26(4), 588-609, 2019

HYBRID MODEL TO DESIGN AN AGRO-FOOD DISTRIBUTION NETWORK


CONSIDERING FOOD QUALITY
Isidro Jesús González-Hernández1,2,*, Rafael Granillo-Macías1, José Luis Martínez-Flores2, Diana Sánchez-Partida2,
and Damián Emilio Gibaja-Romero2
1
Department of Industrial Engineering
Autonomous University of Hidalgo State
Hidalgo, México
*
Corresponding author’s e-mail: igonzalez@uaeh.edu.mx
2
Department of Logistics and Supply Chain Management
Popular Autonomous University of the State of Puebla
Puebla, México

In the agro-food supply chain, the client’s satisfaction is mainly reflected on food freshness in such a way that quality has a
direct relationship with the selling price. This research work suggests a hybrid model to design a food distribution network
that maximizes the farmers’ profits depending on the delivered quality by using a mixed integer linear programming model
and a discrete-event simulation model under specific stochastic parameters. The main contribution of this paper is a model
that evaluates food quality with a discrete and stochastic approach along the whole supply chain. The problem involves only
one product (prickly pear), 93 farmers, 53 distribution centers, and 27 markets. The results determined that the amount of
prickly pear from the 93 farmers must be sent to 29 distribution centers to satisfy the market demand. The delivered quality
of the fruit at the market was in an interval from 84% and 94%.

Keywords: agro-food supply chain; optimization; simulation; hybrid modeling; food quality

(Received on August 9, 2018; Accepted on June 29, 2019)

1. INTRODUCTION

The commerce of fruits and vegetables is one of the most dynamic and essential areas of the world’s economy. Food quality,
safety, and delivery times are the main challenges for those involved in the agro-food supply chain as the client’s satisfaction
is mainly reflected in food quality. Fruits and vegetables usually have a short life cycle and get deteriorated quickly, meaning
that their quality depends on time. Therefore, it is essential to distribute and deliver fruits and vegetables immediately after
they are harvested to limit their degradation. Besides, quality has a direct relationship with the selling price (Wang et al.,
2016; Zhu and Lee, 2018).
At the same time, there is a concern about food loss and waste. Jan et al. (2013) mention that annually the total
agricultural production for food and non-food uses is of 6 gigatonnes, while food waste is of 1.6 gigatonnes, meaning that
around 27% and 30% of food is wasted. It may be due to problems in harvesting, storage, packaging, transportation,
infrastructure or market, and price mechanisms, as well as institutional and legal standards (FAO, 2018).
Miranda-Ackerman et al. (2014) mention that in some developing countries there is an administrative and technological
gap in agro-food supply chains, which leads to economic losses due to low performance of productivity, efficiency, and
quality. Particularly in Mexico, 14 million tonnes of food are lost each year in post-harvesting and distribution processes, this
because of inappropriate transportation and storage conditions, meaning that there are deficiencies in the cold chain or the
cold chain merely is not applied, besides a bad supply and demand forecast (CEC, 2017).
These inefficiencies in the agro-food supply chain result in a loss of capital, work invested, and resources associated
with production and distribution (Jonkman et al., 2019). They also significantly contribute to global warming as the
cultivation, packaging, storage, and transportation processes of the supply chain produce greenhouse gases (GHG) (Manzini
et al., 2015). That way, the economic and environmental performance of the agro-food supply chain is strongly related to the
supply chain configuration. Therefore, reassessing the configuration of the agro-food supply chain can improve its economic
and environmental performance, and address inefficiencies in the chain to improve its sustainability (Su).
However, determining the optimal configuration of the agro-food supply chain is a complex problem, since it is different
from other supply chains. The fundamental difference lies in stochastic factors that affect farmers' crop yields, demand,

ISSN 1943-670X  INTERNATIONAL JOURNAL OF INDUSTRIAL ENGINEERING


González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

quality and shelf life of food, seasonality in production, as well as considering the specific requirements of food for
transportation and storage.
The strategic decision making for agro-food supply chain network design considers the number, type and location of
the facilities at different echelons (agricultural fields, distribution centers, production plants, retailers, and markets or
customers), facilities capacity (crop yield, storage, and transport), the food flow or distribution on the logistics network
between facilities, and stochastic factors related to this type of supply chain. The purpose of designing or redesigning a
distribution network is to meet the objective of minimizing total logistics costs, delivery times, inventory levels, food loss
and waste, GHG emissions, and on the other hand maximizing food quality, customer service level, stakeholder profits,
among others.
For example, Jahani et al. (2018) present an analytical approach to redesign a supply chain network with the objective
of maximizing the total profitability of the chain, in their literature research the authors mention that considering stochastic
demand accompanied by stochastic price, and especially their correlation in the market, has not been widely analyzed. On
the other hand, the work presented by Sazvar et al. (2018) mentions that the majority of the studies carried out in the field of
food supply chains have ignored the food shelf life or deterioration (quality). In this context, this study aimed to develop and
analyze a hybrid model for a distribution network design of prickly pear supply chain in Mexico. Where stochastic factors
such as food quality (food sale price depends on the quality delivered), crop yield, storage times, delivery times, market
demand, and production costs are integrated to maximize the farmers’ profits.
The current characteristics of the prickly pear supply chain, especially the number, location, and crops yields; number
and location of distribution centers; number, location, and market demands, were considered. Then, as part of the hybrid
model, a mathematical model of mixed integer linear programming (MILP) is developed to optimize a deterministic scenario,
the candidate solution obtained from the MILP is simulated from a discrete-event simulation (DES) approach to incorporate
the stochastic factors mentioned in the previous paragraph. The difference between the deterministic objective function and
the objective value of the simulation is incorporated again into the MILP model. Solutions that were previously obtained and
simulated are evaluated by taking into account the impact of stochastic factors on the objective. The process iterates until an
objective value of a previously simulated solution is found to be optimal in the current MILP formulation.
This work contributes to the literature by presenting a hybrid model (optimization-simulation). This model develops
and incorporates a model that evaluates the food quality from a discrete and stochastic approach since selling price of the
prickly pear is in function of the quality delivered, and the quality of the prickly pear depends on stochastic factors such as
the delivery time and the temperature at which the prickly pear was exposed. Also, other stochastic factors are incorporated
such as crop yield, storage times, delivery times, market demand, and production costs.
The remainder of the paper is organized as follows: a literature review is provided in Section 2, where studies that are
related to optimization and simulation of the agro-food supply chain are mentioned, in addition to considering the models
that have been applied to evaluate the food quality in the supply chain. Section 3 presents a case study that includes a
description of the problem addressed. The model to evaluate the food quality in a discrete and stochastic way is included.
The MILP model is also detailed for the deterministic case. The essential aspects of the simulation are included to incorporate
the stochastic factors of the problem. The hybrid model is described. Section 4 presents the computational results. Finally,
the conclusions are made in Section 5.

2. LITERATURE REVIEW

The performance of the agro-food supply chain strongly impacts the global economy; this means that developing optimization
models to support the decision-making process in the agro-food sector is necessary. These models should contemplate agro-
food supply chain inherent characteristics and sources of uncertainty associated with stochastic factors to provide applicable
and accurate solutions.
Therefore, decisions under uncertainty or stochastic factors are a characteristic of an agro-food supply chain, as
perishable foods are particularly sensitive to environmental conditions, having an impact in the performance of different
logistic operations made throughout this chain (De Keizer et al., 2017). Also, Granillo-Macías et al. (2018) mention that the
design of distribution networks in agriculture also implies the analysis of different conditions related to intrinsic and extrinsic
factors of the food that is distributed.
For example, Esteso et al. (2018) conducted a literature review about the mathematical programming (MP) models that
have been applied to design agro-food supply chains under uncertainty. In their analysis, it is mentioned that the most active
modeling type was MILP (more than 50%), followed by multi-objective integer linear programming, stochastic programming,
fuzzy mathematical programming, and heuristics and metaheuristics. However, in this work it was not found any paper that
used hybrid modeling, that is, optimization-simulation.

589
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

In this sense, we did a literature review where our material collection was carried out in Web of Science using the
following search points, Topic: (agro food supply chain) OR Topic: (food supply chain) AND Topic: (optimization) AND
Topic: (simulation). Refined by: Publication years: (2012-2019). In the following paragraphs, we first describe the works that
used mathematical programming models, and then we described the works that used the hybrid approach, that is,
optimization-simulation.
With regard to supply chain network design (SCND), Jonkman et al. (2017) propose a MILP model to determine the
number, capacity, facility locations (FL), the process design (facility type), and how to allocate supply and demand to these
facilities to maximize the profitability in the food supply chain and reduce GHG emissions. Miranda-Ackerman et al. (2017)
propose a modeling and optimization framework for the supply chain network of the food industry. The authors present a
pure integer non-linear problem with multiple objective functions in order to find the optimal trade-off configuration
considering not only operational or economic criteria but also environmental ones. In order to handle the complexity of the
model structure and its components, a multi-objective genetic algorithm (GA) is proposed to find the Pareto optimal solutions.
Accorsi et al. (2015) addressed a land-network problem merges localized and large-scaled decisions as land-use
allocation and location-allocation problems, as well as production planning and distribution (PPD) problems in an agro-food
network. A linear programming (LP) model optimizes infrastructure, agriculture, and logistics costs and also balances carbon
emissions within the agro-food ecosystem. Tsao (2013) presents a non-linear programming model to design a distribution
network of fresh foods integrating the location of facilities through the assignment of optimal inventory levels that allows
keeping foods’ freshness from the farmers to the end clients. Other authors such as Petig et al. (2018) and Sutopo et al. (2013)
also addressed the supply chain network design using MILP.
Another critical aspect in agro-food supply chain management is the inventory policy and determining optimal routes
for vehicles. For example, Tirkolaee et al. (2017) propose a MILP robust model to optimize the multi-trip heterogeneous
vehicle routing (VR) problem of perishable products under uncertainty of demand parameter. The authors use CPLEX solver
in the GAMS optimization software in order to validate the model by solving some random instances. Their model minimizes
total costs that include vehicles usage costs, total traveled distance, earliness and tardiness costs of services. Wang et al.
(2016) tackle a multi-objective problem for the vehicle routing with time windows that deals with food perishability, these
authors apply a two-phase heuristic algorithm based on a variable neighborhood search algorithm (VNSA) and a GA to
minimize the total costs and maximize the freshness of the delivered products.
Aghighi and Malmir (2016) addressed a location routing inventory (LRI) problem. The goals of the proposed model
were to choose, locate, and allocate a set of distribution centers, schedule vehicle routes, and determine distribution center
capacity and reorder point to minimize total costs. The authors used chance-constrained programming (CCP), and the queuing
theory, as well as the simulated annealing heuristic (SAH) in order to solve the model. On the other hand, Micale and La
Scalia (2018) analyzed the performance of a perishable food warehouse considering the effects of the deterioration rate. Two
alternative policies, FIFO and Least Shelf-life First Out (LSFO) were compared concerning their effects on the product quality
in terms of remaining shelf life, as well as to the impact on the economic performance of the system. Authors used RF
technologies to establish a shelf-life based prediction model. Besides, they used an equation to determine the sale price of the
product, but this equation is only based on time.
The topic of production planning and scheduling (PPS) in the agro-food supply chain is addressed by Sazvar et al.
(2018), the authors developed a multi-objective mixed integer non-linear programming (MINLP) model for minimizing the
total cost, reducing GHG emissions, and maximizing social health. Next, by applying some linearization techniques, the
developed model is converted into a MILP model. Finally, Augmented Ɛ-constraint method is applied to solve the proposed
multi-objective model.
Nguyen et al. (2014) formulate a stochastic dynamic programming (SDP) model for a shipments consolidation (SHC)
problem with multiple providers and a consolidation center that delivers perishable products to an only destination. Also,
there is a hard time constraint for the product’s stay in inventory at the consolidation center.
Hybrid approaches based on simulation have been proposed, having as objective to include stochastic factors within an
optimization model (Pourya and Kyoung, 2016), with the advantage that with these models an optimal solution is analyzed
under stochastic conditions (Ge et al., 2016; Almeder et al., 2009).
For example, Allaoui et al. (2018) propose an integrated two-stage hybrid approach for sustainable agro-food supply
chain network design. The first step is to select a partner through a hybrid multi-criteria decision-making based on the analytic
hierarchy process (AHP) method and the ordered weighted averaging (OWA) aggregation method. In stage two, a multi-
objective programming (MOP) model, encompassing all three dimensions of sustainability, is formulated to design the agro-
food supply chain network and to identify the optimal decisions in order to minimize the total costs of the supply chain. Zhao
et al. (2018) present a study to redesign the agro-product supply chain to improve its eco-efficiency by considering the
associated agro-waste flowing into a bioenergy enterprise for energy production. Two operational scenarios are established:
the first one assumes that all waste flows into the enterprise, whereas the second one only considers the inflow of agro-waste
590
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

produced by farmers and the wholesale market. They used system dynamics simulation (SDS) to select the optimal scenario
to enhance eco-efficiency for the proposed supply chain.
Another topic that has been addressed with simulation optimization is the inventory policy (IP). Ambekar and Kapoor
(2019) present an iterative process, simulation optimization via the GA, to optimize the inventory policies for the distribution
stage of the supply chain (public food distribution system) to minimize the total distribution cost. A two-stage modeling
approach was adopted for this purpose. In the first stage, a simulation model was developed for a periodic review-based stock
policy with appropriate assumptions. The objective here was to minimize the total supply chain cost. In the second stage, a
GA-based optimization approach was used. The GA implementation is based on simulation for a given set of policy
parameters. The GA aims to identify an optimal set of policy parameters of the system.
Noordhoek et al. (2018) propose a simulation-based optimization framework as the best method in terms of solution
quality to optimize the (s, S) inventory policies of a multi-level food distribution network with deterministic lead times,
backorder, and fill rate constraints. The authors used nested bisection search (NBS), scatter search metaheuristic (SSM), and
DES. Beshara et al. (2012) addressed an inventory replenishment problem of an agro-food supply chain of highly perishable
products. The objective is to improve a set of performance measures by developing a simulation model that helps in evaluating
and analyzing the performance of the chain. The supply chain network consists of two single node echelons representing
production and distribution. The production echelon is located in Egypt while the distribution echelon is in Holland.
On the other hand, Sel et al. (2017) develop an optimization-simulation model for the yogurt production planning and
scheduling problem considering the characteristics such as the predefined production sequence, the uncertain lifetime of
intermediate products and the make-and-pack process of yogurt production. The authors used a MILP model. After, a
simulation model is introduced to evaluate the proposed production schedule. The simulation aims to calculate production
waste. Ge et al. (2016) present a hybrid model of optimization-simulation for the supply chain of wheat in Canada, to identify
profitable strategies for different varieties of wheat. This model solves a base scenario, and it later performs a sensitivity
analysis of the critical variables that influence the strategies to evaluate the quality of wheat. The numerical simulation is
used as a complementing tool to integrate complex characteristics of the system in the optimization model for risk mitigation
(RM).
In Table 1 it is presented a summary of the main characteristics of the works cited in the review of the literature of
which the food quality evaluation and stochastic factors are two aspects that we are considering important in this work for
the design of a distribution network. The authors who evaluate the quality or freshness of the food in their works are Micale
and La Scalia (2018); Sel et al. (2017); Sutopo et al. (2013); Wang et al. (2016). However, in their models, food quality is
only a function of time; they do not consider the temperature variable to evaluate product quality. Another author such as
Jonkman et al. (2017) considers the food quality, but quality appears as a parameter in the model. In this sense, one of our
proposals is to incorporate time and temperature variables in a model that evaluates food quality in a discrete and stochastic
way, that is, the product quality will be evaluated in the main supply chain stages considering stochastic data of time and
temperature in a simulation model. Finally, only Jonkman et al. (2017) and Sel et al. (2017) consider stochastic factors in
their work and a model to evaluate food quality.

Table 1. Related recent papers in the context of agro-food supply chain optimization

Author(s) Problem Stochastic Food Objective function Solution


factors quality approach
(Ambekar and Kapoor, 2019) IP ✓ MIN Costs S, GA
(Allaoui et al., 2018) SCND MIN Costs AHP, MOP
(Petig et al., 2018) FL MAX Profits MILP
(Sazvar et al., 2018) PPS, Su MIN Costs MINLP
MIN GHG
MAX Social Health
(Micale and La Scalia, 2018) IP ✓ Minimize Food RF
Waste Technologies
(Noordhoek et al., 2018) IP ✓ MIN Costs NBS, SSM,
DES
(Zhao et al., 2018) SCND MIN GHG SDS

591
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

(Jonkman et al., 2017) SCND ✓ ✓ MAX Profits MILP


MIN GHG

(Miranda-Ackerman et al., SCND MAX Net Present MINLP, GA


2017) Value
MIN GHG
MIN Costs
(Sel et al., 2017) PPS ✓ ✓ MIN Makespan MILP,
Simulation
(Tirkolaee et al., 2017) VR ✓ MIN Costs MILP

(Aghighi and Malmir, 2016) LRI ✓ MIN Costs CCP, SAH

(Ge et al., 2016) RM ✓ MIN Risks and Costs MP,


Simulation

(Wang et al., 2016) VR ✓ MIN Costs MOP, VNSA,


GA
(Accorsi et al., 2015) PPD, Su MIN Costs LP
MIN GHG

(Nguyen et al., 2014) SHC ✓ MIN Costs SDP

(Sutopo et al., 2013) SCND ✓ MAX Profits MILP

(Tsao, 2013) SCND MAX Profits MINLP

(Beshara et al., 2012) IP ✓ Performance Simulation


Measuring

3. METHOD

3.1 Case study

This research study analyses the supply chain of prickly pear, of which Mexico is the most significant producer worldwide.
According to SIAP (2018) data, Mexico has 20,000 farmers dedicated to growing this product. They harvest an average of
352,000 tonnes per year in 48,000 hectares available for sowing.
One problem of this sector is that the national production of prickly pear is frequently scattered throughout climates and
lands with average surfaces of five hectares per farmer, so distribution and commercialization operations become crucial for
the profitability of this agroindustrial sector (Ramírez-Abarca et al., 2015; Pinedo-Espinoza et al., 2010). The primary
producers and merchandisers of this fruit are in the central zone of the country, and they supply different national markets
(Ayllon et al., 2015).
Additionally, as this product is perishable, the time between harvest and the delivery to the client is a factor that has a
direct impact on the quality of the delivered fruit and consequently on the selling price reducing profits along this supply
chain mainly affecting the farmers.
Besides, as a result from the declining conditions of quality and the seasonality itself of prickly pear’s production in the
months from June to November, this supply chain includes too many intermediaries for distribution, which results in low
costs for farmers and high costs for consumers (Ramírez-Abarca et al., 2015).
Based on this problem, it was suggested a model that allows optimizing a distribution network for prickly pear with the
maximum profit possible considering the costs related to cultivation, harvest, transportation, distribution, and storage, besides
incorporating stochastic factors related with quality degradation of the fruit.
The scope of this case study considered 93 of the leading farmers in the country that harvest prickly pears in the State
of Mexico, Zacatecas, Puebla, Hidalgo, San Luis Potosí and Guanajuato (México Produce, 2016). It was proposed that the
592
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

product obtained from the farmers is sent in a consolidated way through the available infrastructure for storage to avoid
intermediaries. In such way that, based on data from (ASERCA, 2017) 53 distribution centers located in the same producing
regions were selected; the consolidated load is sent to the clients from the distribution centers, for that, 27 potential consumer
markets of prickly pear were located along the country (Figure 1).

Figure 1. Distribution of farmers, distribution centers and markets

3.2 Model to evaluate the quality of prickly pear

Food quality can be measured by color, presence or absence of certain flavor compounds, presence or absence of certain
microorganisms, texture, vitamins, protein composition, among others. The kinetic equation is the model used to determine
the quality of food from a sensory perspective (Equation 1). This equation describes food degradation based on one or several
characteristics such as smell, aromas, flavor, volume, and weight (Van Boekel, 2008).

𝑑𝑄(𝑡)
= −𝑘𝑄0 (𝑡)𝑛 (1)
𝑑𝑡

Where 𝑄 is the quality attribute to be quantified, 𝑄0 is the initial quality of the food, 𝑡 is the time, 𝑘 is the reaction rate
constant (rate constant of quality loss) and, 𝑛 is the reaction kinetic order. Ling et al. (2015) mention that the kinetic reaction
of food, generally follows a reaction order 𝑛 = 0, 𝑛 = 1 o 𝑛 = 2, according to the intrinsic characteristics to be evaluated.
The solutions of the kinetic equation for the former rates are shown in Table 2.

Table 2. The solution of the kinetic equation for the three reaction orders

𝒏=𝟎 𝒏=𝟏 𝒏=𝟐


𝑑𝑄(𝑡)
𝑑𝑄(𝑡) 𝑑𝑄(𝑡) = −𝑘𝑄(𝑡)2
= −𝑘𝑄(𝑡)0 = −𝑘𝑄(𝑡) 𝑑𝑡
𝑑𝑡 𝑑𝑡
𝑄0
𝑄(𝑡) = 𝑄0 − 𝑘𝑡 𝑄(𝑡) = 𝑄0 𝑒 −𝑘𝑡 𝑄(𝑡) =
𝑄0 ∙ 𝑘 ∙ 𝑡 + 1

This way, the behavior of quality degradation for 𝑛 = 0, has a linear adjustment. When 𝑛 = 1, is adjusted to an
exponential behavior, and for 𝑛 = 2, the adjustment is hyperbolic, as shown in Figure 2.

593
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

100

90 n=0 n=1 n=2


80

70

60

Quality
50

40

30

20

10

0
0 2 4 6 8 10 12 14 16 18 20

Time
Figure 2. Quality degradation for the three reaction orders

Temperature is another variable that influences food degradation speed, and it is considered as the primary cause that
affects its quality and shelf life. Ling et al. (2015) mention that the Arrhenius model is the most commonly used method to
relate the chemical reaction rate with temperature changes 𝑇 by the reaction rate constant 𝑘, as shown in Equation 2.

𝐸𝐴
− (2)
𝑘[𝑇(𝑡)] = 𝑘𝐴 ∗ 𝑒 𝑅∙𝑇(𝑡) ,

where 𝑘[𝑇(𝑡)] is the reaction rate constant, this variable is the same 𝑘 of Equation 1, 𝑘𝐴 is the pre-exponential factor or
constant rate, 𝐸𝐴 is the activation energy (J/mol), 𝑅 is the ideal constant of gases (8.314 J/mol·K) and 𝑇(𝑡) is the absolute
temperature in Kelvin scale. Therefore, the model to evaluate food quality is expressed as shown in Table 3.

Table 3. The solution of the kinetic equation

Order of the kinetic reaction Solution

𝐸𝐴

𝑛=0 𝑄 [𝑇(𝑡)] = 𝑄0 − [𝑘𝐴 ∗ 𝑒 𝑅∙𝑇(𝑡) ]∗ 𝑡

𝐸𝐴

𝑛=1 −[𝑘𝐴 ∗𝑒 𝑅∙𝑇(𝑡) ]∗𝑡
𝑄[𝑇(𝑡)] = 𝑄0 𝑒

𝑄0
𝑄[𝑇(𝑡)] = 𝐸𝐴
𝑛=2 −
𝑄0 ∗ [𝑘𝐴 ∗ 𝑒 𝑅∙𝑇(𝑡) ]∗ 𝑡+1

Because fruits and vegetables generally have a short life cycle, food deteriorates rapidly, that is, the quality or shelf life
depends on the time and temperature at which food is exposed. In such a way that, the agro-food supply chain has become
increasingly complex because all or some parts of the harvest, distribution, storage or processing are carried out in multiple
locations around the world, therefore, on average, the distance from the farm to the table has increased. This increase in
distance has a direct relationship with delivery times, food freshness (quality), customer satisfaction, and the price to be
charged. In this context, it is essential to measure the quality of the food during prolonged periods, at different temperatures
in order to obtain precise predictions that allow determining the quality of the food in the supply chain. To obtain the model

594
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

that evaluates the quality of prickly pear in a discrete and stochastic way (stochastic times and temperatures), the kinetic
equation (1) and the Arrhenius equation (2) will be used.
The work of Nunes et al. (2014) was taken as an example to analyze how the quality of prickly pear behaves. For this
study, the quality of the prickly pear was evaluated from a sensory perspective (color and texture) at different temperatures
(0, 10, 20 and 30 °C), during 30 days, as shown in Figure 3. Besides, it was considered that 100% is the top quality and 0 is
not suitable for consumption.

Quality (100 – 0 %)

30
Temperature ( C)

20

10

0 5 10 15 20 25 30
Time (days)
Figure 3. Quality of prickly pear at different storage temperatures

To determine the kinetic reaction rate (𝑛 = 0 or 1) at which the quality behavior of prickly pear is adjusted, it was used
the method of data adjustment through a linear regression for 𝑛 = 0 and an exponential regression for 𝑛 = 1, for each
temperature (0, 10, 20 y 30 °C). The estimator that helps to determine at which reaction rate the data is adjusted, is the
coefficient of determination 𝑅2 ; this estimator is a measure that relates the existing rate between the dependent variable
(quality) and the independent variable (temperature). A value close to 0 indicates that there is no relation between the
dependent and independent variable, while a value close to 1 indicates a perfect adjustment of the model with the experimental
data. According to the data of 𝑅2 , the behavior of the prickly pear’s quality is adjusted at a reaction rate 𝑛 = 0.

Once the reaction rate was determined, then the values of the Arrhenius equation were obtained for a temperature
interval of 0 to 30 °C. For this, it was used the linear regression method by using the slope values (𝑚) of each one of the
temperatures for 𝑛 = 0, it is worth mentioning that in this case, the slope value represents 𝑘. The obtained data of 𝑘 with 𝑛 =
0 are shown in Table 4.

Table 4. Slope values for the four temperatures

°C 𝒎=𝒌
0 0.00025
10 0.00055
20 0.00085
30 0.0019

Considering the Arrhenius equation in a logarithmic way 𝑙𝑛𝑘[𝑇(𝑡)] = 𝑙𝑛𝑘𝐴 − 𝐸𝐴 /𝑅 ∗ 1/𝑇(𝑡), the temperature at
Kelvin scale, and the changes of variable 𝑦 = 𝑙𝑛𝑘[𝑇(𝑡)] y 𝑥 = 1/𝑇(𝑡). Using the linear regression method with 𝑦 = 𝑏 +
𝑚𝑥, where 𝑏 = 𝑙𝑛𝑘𝐴 and 𝑚 = 𝐸𝐴 /𝑅, it is obtained the solution 𝑦 = 11.455 − 5388.7 𝑥, which graph is shown in Figure 4.

595
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

0.0032 0.0033 0.0034 0.0035 0.0036 0.0037


0

-1

-2

-3

Ln(m = kA )
-4

-5

-6

-7

-8

-9
1/T

Figure 4. Linear regression to obtain Arrhenius equation

By performing algebraic operations, where 𝑙𝑛𝑘𝐴 = 11.455 → 𝑘𝐴 = 𝑒 (11.455) = 94372.028 and 𝐸𝐴 /𝑅 = 5388.7 →
𝐸𝐴 = (5388.7)(0.00831) = 44.780097. The Arrhenius Equation (3) is obtained with the specific values for the prickly
pear.

44.780097
− (3)
𝑘[𝑇(𝑡)] = 94372.0285 ∗ 𝑒 (0.00831)∙𝑇(𝑡)

This way, the model to determine prickly pear’s quality is the following:

44.780097

𝑄[𝑇(𝑡)] = 100 − [94372.0285 ∗ 𝑒 (0.00831)∙𝑇(𝑡) ]∗ 𝑡 (4)

Finally, considering that temperature varies along the distribution route, model (4) can be evaluated in a discrete and
stochastic way being expressed as follows:

44.780097
− 𝐹(𝑥)
(0.00831)(𝑇 ) 𝐹(𝑥)
𝑄[𝑇(𝑡∆ )] = 100 − ∑𝑚
𝑖=1 (94372.0285 ∗𝑒 (𝑡∆ )
) 𝑡∆ , (5)

where 𝑡∆ is the elapsed time to carry out a process in the supply chain. 𝐹(𝑥) is a distribution of probability for time and
𝐹(𝑥)
temperature. Therefore, 𝑇(𝑡∆) is the temperature in Kelvin scale that will take values according to a distribution of probability
and it will depend on temperatures at which a process is developed at the moment of evaluating the food quality. On the other
𝐹(𝑥)
hand, 𝑡∆ is the elapsed time to carry out a process that depends on other distribution of probability.

3.3 Optimization model

The suggested MILP model was formulated considering maximizing the farmers’ profits based on the costs and the demand
requirements of different consumer markets’ demands. The results of this model show in a deterministic way the best
assignment of the farmers’ harvest–distribution centers and distribution centers–markets, indicating as well which distribution
centers must be open for supply at the lowest cost possible along the supply chain.

Indices:

𝐼, 𝐽, 𝐾, 𝑁 ∈ ℤ+ number of different farmers, distribution centers, markets and candidate


solutions, respectively.
596
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

𝑖, 𝑗, 𝑘, 𝑛 indices covered by different farmers, distribution centers, markets and candidate


solutions, respectively.

Variables of decision:

𝑆𝐷𝑖𝑗 amount of product to be transported from the farmer 𝑖 to the distribution center 𝑗,
𝑆𝐷𝑗𝑘 amount of product to be transported from the distribution center 𝑗 to the market 𝑘,
𝑌𝑗 1 if the distribution center j opens, 0 if not,
𝑍𝑛 a binary variable that incorporates the cost of stochastic factors to the candidate solution
𝑛.

Optimization parameters:

𝑃𝑘 the selling price of prickly pear at the market 𝑘,


𝑇𝑖𝑗 transportation cost from the farmer 𝑖 to the distribution center 𝑗,
𝑇𝑗𝑘 transportation cost from the distribution center 𝑗 to the market 𝑘,
𝑅𝑖 crop yield of the farmers 𝑖,
𝐶𝑖 cost of the supplies for the farmers’ sowing 𝑖,
𝐹𝑖 cost of the internal factors for the farmers’ sowing 𝑖,
𝑠𝑢𝑚𝑖 the available offer of the farmer 𝑖,
𝑐𝑎𝑝𝑗 the storage capacity of the distribution center 𝑗,
𝑑𝑒𝑚𝑘 market demand 𝑘,
𝑂𝐶𝑗 opening cost of the distribution center 𝑗,
𝑆𝑛 cost of stochastic factors for the scenario in iteration 𝑛,
𝑈𝑛𝑗 binary parameter (1, if binary variable 𝑌𝑗 in candidate solution 𝑛 is 1; 0, otherwise),
𝑄𝑚𝑖𝑛 minimum total cost obtained so far from the simulations in previous iterations,
𝑀 large number.

In the objective function (6), the farmers (𝐼) are located in the productive zones of prickly pears in Puebla, Zacatecas,
San Luis Potosí, State of Mexico, Hidalgo and Guanajuato; the distribution centers (𝐽) that were considered for the network
design depend on the available infrastructure according to (ASERCA, 2017) and the markets (𝐾) of potential consumption
within the country; they are located in Aguascalientes, Baja California, Campeche, Coahuila, Colima, Chiapas, Mexico City,
Durango, Guanajuato, Guadalajara, Michoacán, Morelos, Tepic, Nuevo León, Oaxaca, Querétaro, Quintana Roo,
Tamaulipas, Veracruz, and Yucatán.

Maximize

𝐽 𝐾 𝐼 𝐽 𝐼 𝐽 𝐼 𝐽
𝐶𝑖 𝑆𝐷𝑖𝑗 𝐹𝑖 𝑆𝐷𝑖𝑗
𝑍 = ∑ ∑ 𝑃𝑘 𝑆𝐷𝑗𝑘 − ∑ ∑ −∑∑ − ∑ ∑ 𝑇𝑖𝑗 𝑆𝐷𝑖𝑗
𝑅𝑖 𝑅𝑖
𝑗=1 𝑘=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1 𝑖=1 𝑗=1
𝐽 𝐾 𝐽 𝑁 (6)
− ∑ ∑ 𝑇𝑗𝑘 𝑆𝐷𝑗𝑘 − ∑ 𝑂𝐶𝑗 𝑌𝑗 + ∑ 𝑆𝑛 𝑍𝑛
𝑗=1 𝑘=1 𝑗=1 𝑛=1

The decision variables 𝑆𝐷𝑖𝑗 and 𝑆𝐷𝑗𝑘 indicate the number of tonnes to be transported from the farmer 𝑖 to the distribution
center 𝑗 and from the distribution center 𝑗 to the consumer market 𝑘. 𝑃𝑘 indicates the selling price of the prickly pear at the
market 𝑘, for the MILP model the prickly pear sale price is considered as a deterministic value, that is, product quality is not
considered, only a sale price is established. It is up to the simulation stage where the model (5) and stochastic factors are
incorporated to evaluate the prickly pear quality and establish the sale price based on the delivered quality. 𝑌𝑗 is a binary
variable that indicates which distribution centers must be used and which must not with the purpose of optimizing the
distribution network. Therefore 𝑂𝐶𝑗 is the fixed cost of the opening of the distribution center 𝑗.

597
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

The production costs ($/h) are identified as 𝐶𝑖 costs that include supplies for sowing (fertilizers, fungicides, herbicides,
insecticides, seed or plant, fuels, and services) and 𝐹𝑖 are the costs associated with internal factors for sowing such as manual
works, asset loans, and several materials. 𝑅𝑖 is a parameter that indicates the crop yield (tonne/h) obtained by certain farmer
𝑖. The parameters 𝑇𝑖𝑗 and 𝑇𝑗𝑘 in (6) include the transportation costs ($/tonne) from the farmer 𝑖 to the distribution center 𝑗 and
from the distribution center 𝑗 to the market 𝑘.
Additionally, 𝑆𝑛 indicates the impact of stochastic factors found in the simulation of the candidate solution 𝑛. 𝑍𝑛
represents a binary variable that incorporates the cost of stochastic factors for the candidate solution 𝑛. A candidate solution
is defined as an optimal deterministic scenario previously obtained under certain parameters. That is, when the MILP model
obtains a new candidate solution, it is simulated and the stochastic factors of interest for the problem are incorporated. The
importance of the stochastic factors in this model lies in the fact that the sale price of the prickly pear is directly related to
food quality and food quality is directly related to stochastic factors such as time and temperature at which food is exposed.
In such a way that in the simulation, these stochastic factors are incorporated to evaluate the prickly pear quality and determine
the sale price. In the simulation, several replicas are made to obtain the objective value (𝑍𝑠𝑖𝑚 ). The difference between
deterministic objective function value (𝑍) and the simulation objective value (𝑍𝑠𝑖𝑚 ) is considered as the cost or impact of the
stochastic factors, where 𝑆𝑛 = 𝑍𝑠𝑖𝑚 − 𝑍.
So the objective function consists on the sale of the product in the market 𝐾 minus the production costs, the costs
associated with internal factors such as asset loans, the transport costs of 𝑖 − 𝑗 and 𝑗 − 𝑘, the cost of the distribution center
opening 𝑗, and the cost of stochastic factors 𝑆𝑛 .
The restrictions for this model are: the expression (7) indicates the amount sent from the distribution center 𝑗 to the
market 𝑘 that it must be higher or equal to the demand in 𝑘.
𝐽

∑ 𝑆𝐷𝑗𝑘 ≥ 𝑑𝑒𝑚𝑘 for each 𝑘 = 1,2, … , 𝐾 (7)


𝑗=1

Restriction (8) guarantees that the production of prickly pear that is sent from the farmer 𝑖 to the distribution center 𝑗 is
not higher than the available harvest of 𝑖.
𝐽

∑ 𝑆𝐷𝑖𝑗 ≤ 𝑠𝑢𝑚𝑖 for each 𝑖 = 1,2, … , 𝐼 (8)


𝑗=1

Restrictions (9) and (10) relate the amount sent 𝑆𝐷𝑖𝑗 and 𝑆𝐷𝑗𝑘 with the available capacity at the distribution center 𝑗.

∑ 𝑆𝐷𝑖𝑗 ≤ 𝑐𝑎𝑝𝑗 for each 𝑗 = 1,2, … , 𝐽 (9)


𝑖=1

∑ 𝑆𝐷𝑗𝑘 ≤ 𝑌𝑗 𝑐𝑎𝑝𝑗 for each 𝑗 = 1,2, … , 𝐽 (10)


𝑘=1

The flow (inflows-outflows) at the consolidation center is expressed in (11) as a balance equation.
𝐼 𝐾

∑ 𝑆𝐷𝑖𝑗 = ∑ 𝑆𝐷𝑗𝑘 for each 𝑗 = 1,2, … , 𝐽 (11)


𝑖=1 𝑘=1

Restrictions (12) and (13) are similar to the proposals by Acar et al. (2009) to incorporate the cost of the stochastic
factors coming from a simulation model. Therefore, equations (12), (13) and (14) are incorporated into the MILP model after
simulating a candidate solution.

598
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

∑(2𝑈𝑛𝑗 𝑌𝑗 − 𝑈𝑛𝑗 − 𝑌𝑗 ) ≤ 𝑍𝑛 − 1 for each 𝑛 = 1,2, … , 𝑁 (12)


𝑗=1

∑(2𝑈𝑛𝑗 𝑌𝑗 − 𝑈𝑛𝑗 − 𝑌𝑗 ) ≥ 𝑀(𝑍𝑛 − 1) for each 𝑛 = 1,2, … , 𝑁 (13)


𝑗=1

Restriction (14) is included in the model to establish a lower bound that is obtained in the simulation of the previous
candidate solution.

𝑍 ≥ 𝑄𝑚𝑖𝑛 (14)

Finally, (15-18) guarantee the non-negativity of variables and their type.

𝑆𝐷𝑖𝑗 ≥ 0 for each 𝑖 = 1,2, … , 𝐼, 𝑗 = 1,2, … , 𝐽 (15)

𝑆𝐷𝑗𝑘 ≥ 0 for each 𝑗 = 1,2, … , 𝐽, 𝑘 = 1,2, … , 𝐾 (16)

𝑌𝑗 ∈ {0,1} for each 𝑗 = 1,2, … , 𝐽 (17)

𝑍𝑛 ∈ {0,1} for each 𝑛 = 1,2, … , 𝑁 (18)

3.4 Simulation

One of the main challenges of managers and researchers of the agro-food supply chain is to include stochastic factors and
dynamics of this type of chain in optimization models. The DES has the advantage of modeling dynamic and stochastic
aspects that mathematical programming does not take into account (González-Hernández et al., 2019). Simulation model
represents the second stage of the hybrid model where stochastic factors are considered as food quality, crop yield, storage
times, delivery times, market demand, and production costs. The objective of this simulation model is to evaluate the
reliability of the design of the distribution network previously obtained in the MILP model.
The simulation was modeled using the Promodel software. Figure 5 shows the structure of the simulation model. Prickly
pear distribution starts with the farmer 𝑖, who grows ℎ𝑐 hectares of the food, the crop yield 𝑅𝑖 is obtained from a probability
distribution function. Then, the product is stored in the farmer's warehouse for 𝑡 time, at the end of the storage, the food
quality is evaluated using Equation (5), considering that, the quality is a function of time 𝑡 and temperature 𝑇. After the
product is sent to the distribution center where it requires 𝑡 time to get to the location, the product is stored for 𝑡 time. When
the product arrives and leaves the distribution center, the prickly pear quality is evaluated. Finally, the product is sent to the
market where it requires 𝑡 time to reach the location, and prickly pear quality is evaluated to determine the sale price. The
amount of food sent from the distribution center to the market is based on the demand 𝑑𝑒𝑚𝑘 . The time, temperature, and
demand is a probability distribution function in the simulation model. The product distribution (distribution network) in the
simulation is based on the solution previously obtained in the MILP model. Table 5 shows the variables and parameters used
in the simulation.

Excel
interface
Candidate
solution

Input (load data) Farmer DC'S Market


1. Solution obtained from 4. Farmer creates product 8. Evaluating quality 11. Evaluating
Output
MILP 5. Storage 9. Storage quality
13. Objective value
2. Variables 6. Evaluating quality 10. Send product to 12. Calculate sale
3. Parameters 7. Send product to DC's Market price

Figure 5. Structure of the discrete-event simulation model


599
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

Table 5. Inputs and outputs for the simulation model

Stochastic input Deterministic input


Farmer 𝒊
Crop yield ~ 𝑁(𝜇, 𝜎) Transportation cost 𝑖 − 𝑗
Storage times ~ 𝑁(𝜇, 𝜎)
Production costs ~ 𝑁(𝜇, 𝜎)
Food quality (Equation (5)):
time ~ 𝑁(𝜇, 𝜎), Temperature ~ 𝑇(𝑎, 𝑚, 𝑏)
Transportation time ~ 𝑁(𝜇, 𝜎)
Distribution Center 𝒋
Storage times ~ 𝑁(𝜇, 𝜎) Transportation cost 𝑗 − 𝑘
Food quality (Equation (5)): Opening cost of the distribution
time ~ 𝑁(𝜇, 𝜎), Temperature ~ 𝑇(𝑎, 𝑚, 𝑏) center
Transportation time ~ 𝑁(𝜇, 𝜎) Storage capacity
Market 𝒌
Food quality (Equation (5)):
time ~ 𝑁(𝜇, 𝜎), Temperature ~ 𝑇(𝑎, 𝑚, 𝑏)
The sale price of the prickly pear depending on the quality delivered
Product demand ~ 𝑁(𝜇, 𝜎)

Output: objective value (𝑍𝑠𝑖𝑚 )

The simulation output is the income or profits of all network. In such a way that, the objective value 𝑍𝑠𝑖𝑚 of the
simulation is integrated by the prickly pear sale in function of the quality minus the production costs, opening cost of the
selected distribution centers, transportation cost 𝑖 − 𝑗, and transportation cost 𝑗 − 𝑘. Figure 6 shows the total profit of the
entire network for a candidate solution.

Prickly pear sale 204706172.15

Total cost of cultivation 46261456.60

Total cost of opening the DC's 14500.00

Total transportation cost i-j 5789721.02

Total transportation cost j-k 12284952.80

Simulation objective value 𝒁 150401541.69

Figure 6. Simulation of the distribution network

Considering that the prickly pear sale price depends on the delivered quality at the market, prickly pear’s quality was
simulated from farmer-to-market using the model (5). In Figure 7 it can be observed, for example, that the delivered quality
from the farmers to the distribution centers, in a particular case, can reach levels between 97% and 98%, while quality
delivered to the markets gets between 86.5% and 88%. For this specific case, the model that evaluates food quality when the
product arrives at the distribution center and incorporates the stochastic factors of temperature and time is as follows,

600
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

44.780097
− (0.00831)(𝑇(281.15,
𝑄[𝑇(𝑡∆ )] = 100 − ((94372.0285 ∗ 𝑒 283.15, 285.15)) ) (𝑁(2880, 720))) −
(19)
44.780097
− (0.00831)(𝑇(280.15,
((94372.0285 ∗ 𝑒 283.15, 286.15)) ) (𝑁(1440, 720))),

where 𝑇 and 𝑁 is a Triangular and Normal distribution function respectively. It is worth mentioning that the model grows
every time the food quality is evaluated.

Prickly pear quality Prickly pear quality - 95% C.I. Low Prickly pear quality - 95% C.I. High

100.00
99.00
98.00
97.00 Harvest and Storage
96.00
95.00 Transport
Quality

94.00
Storage
93.00
Transport
92.00
91.00
90.00 i j k
89.00
88.00
87.00
86.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00
Days
Figure 7. Quality degradation of a prickly pear through a network

3.5 Hybrid model

The combination of optimization and simulation within a hybrid approach was carried out using a deterministic solution of
the MILP model that was later simulated to include in the objective function the impact of the stochastic factors 𝑆𝑛 such as
quality degradation (stochastic times and temperatures), crop yield, product demand, storage and transportation times.
According to Acar et al. (2009), the difference between the optimization and the simulation value is included again in the
MILP model to formulate a new optimization. This process is executed in an iterative form until finding the network design
that in this case maximizes the farmers’ total profit considering the impact of the stochastic factors coming from the specific
conditions for this supply chain.
In other words, the first stage of the hybrid model consists of obtaining a candidate solution of the MILP model. In the
second stage, the candidate solution is simulated to obtain a 𝑍𝑠𝑖𝑚 . In the third stage, the MILP model is updated by
incorporating the new value of 𝑄𝑚𝑖𝑛 , as well as including the new parameters of 𝑆𝑛 , 𝑍𝑛 , 𝑈𝑛𝑗 and the set of variables. In such
a way that the procedure iterates back to Step 1 until a previously simulated solution is obtained as the optimal solution from
MILP.
Figure 8 shows, in a schematic way, the procedure for the solution, where a MILP model was used to determine in the
first place the network design, this means that it selects the distribution centers 𝑌𝑗 that must be considered for (opening) and
the flows in the amounts to be sent 𝑆𝐷𝑖𝑗 and 𝑆𝐷𝑗𝑘 . Then, this design is evaluated through simulation to calculate the impact
of the stochastic factors 𝑆𝑛 also obtaining a value 𝑄𝑚𝑖𝑛 updating if applicable, the values of 𝑍𝑛 and 𝑈𝑛𝑗 that are incorporated
to evaluate a possible solution in particular. The information obtained from the former stages is useful as feedback about the
optimization model and the network design, the procedure continues evaluating other candidate solutions until finding the
one with the best design under stochastic conditions.

601
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

Problem characterization: Discrete-event simulation


Development of simulation
Complexity, Interactions,
model Select the type of software: Promodel
Dinamycs
Definition problem

Overall project plan

Modeling the input datas (function of


distribution)
Choose type of optimization Incorporate the uncertainty
(MLIP) of the problem Model to evaluate prickly pear quality
Mathematical Define indices, parameters and 𝑚 −
44.780097
𝐹(𝑥)
formulation decision variables (𝑅) 𝑇(𝑡 ) 𝐹(𝑥)
𝑄[𝑇(𝑡∆)] = 100 − ∑ (94372.0285 ∗ 𝑒 ∆ ) 𝑡∆
𝑖=1
Objective function and constraints

Run Model Selected hub to opening


Simulate candidate solution Product flows
max f (x) Prickly pear quality

C (x)

xϵ D Calculate difference
between deterministic cost
and average simulation cost
Value of the objective function
Candidate solution
Network design
Update with uncertainty
impact

Has the solution NO


already
been simulated?

YES

Terminate with optimal


solution

Figure 8. Stages of the hybrid model

The software Matlab, Microsoft Excel, and ProModel were used to solve the hybrid model. First, it was executed the MILP
model by using programming in Matlab and later, through a Microsoft Excel interface, the results obtained by the optimization
are included as inputs in the ProModel simulation model. This process is repeated until finding a distribution network design
that re-optimizes the solution previously obtained minimizing the impact of the stochastic factors between the MILP model
and the simulation.

4. RESULTS AND DISCUSSION

Sazvar et al. (2018), Esteso et al. (2018), and Soto-Silva et al. (2016) mention that practically all optimization models consider
a constant sale price of the food over time without taking into account loss in the product’s value due to quality deterioration.
Also, Soto-Silva et al. (2016) explain that the new generation of decision models for the agro-food supply chain must include
variables such as food quality deterioration or waste, and sale price according to the quality.
In this sense, the superiority of the model herein presented over other optimization models is that this model proposes a
hybrid model that evaluates food quality considering the two main variables that affect food freshness, which is temperature
and time. This model evaluates product quality in a discrete and stochastic way, that is, food quality is evaluated in the main
602
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

stages of the supply chain considering stochastic data of time and temperature in the simulation, since, the food sale price is
based on the delivered quality. Also, other stochastic factors such as crop yield, storage and delivery times, production costs,
and market demand are incorporated into the simulation. The difference between the articles cited which incorporate food
quality and the research work herein is that those works do not incorporate the temperature variable to determine food quality.
The resulting network according to the objective function and restrictions was defined by 6,414 variables of which 53
were of the integer type. In the solution of this problem, three possible scenarios were implemented, which assume variations
in the selling price and the production (cultivation) costs, a recurring situation in this supply chain. In one scenario, the costs
associated with internal factors for sowing as well as the supplies increased 10% considering as well a variation of 10% in
the selling price of the product according to different markets where it is consumed. The experimental scenarios are
summarized in Table 6 below.

Table 6. Experimental scenarios

Experimental factors Variation level


Production costs (𝐶𝑖 + 𝐹𝑖 ) 10% 20% 30% of mean production costs
Sale price (𝑃𝑘 ) 10% 20% 30% of mean sale price

Table 7 shows some of the variations that were made for the analysis of the first scenario (1-10%). Each combination
was optimized in a deterministic way using the MILP model without considering the constraints (12) and (13) that incorporate
stochastic factors, as well as ∑N
n=1 Sn Zn . After obtaining the results of the MILP model for each combination, the candidate
solution (deterministic solution) was chosen as the one with the best gain in its objective function. In this case, combination
8 of Table 7 was selected to simulate the distribution network that resulted from the MILP model.

Table 7. MILP model results (first scenario)

# combinations Selling price Production costs Objective function


% variation (Revenue in millions)
1 2 4 150.69
2 7 4 154.12
3 8 9 155.45
4 6 5 153.69
5 3 6 154.07
6 5 7 156.21
7 6 9 157.67
8 9 3 168.66
9 4 8 158.50
10 3 1 150.12

The deterministic solutions (combination #8 of table 7) were simulated by incorporating stochastic factors such as crop
yield, delivery times, storage times, food quality (time and temperature), market demand, and production costs (cultivation).
Chwif et al. (2013) mention that ten replications must be considered to enhance the precision of the simulation results. In this
case, 20 replications were considered for each scenario. Through an iterative process proposed by Acar et al. (2009), the
candidate solution is simulated and then restrictions (12) and (13) are incorporated as well as ∑N n=1 Sn Zn in the MILP model,
the process of optimizing and simulating continues until there is no better solution than the one previously simulated. For this
first scenario, the results obtained are shown in Table 8.

Table 8. Hybrid model results (first scenario)

Millions Location and number of distribution centers that must be opened


Iteration
MILP Simulation 𝑆𝑛 Guanajuato Hidalgo Mexico Puebla San Luis Zacatecas
1 168.66 142.62 26.04 4 7 4 7 1 6
2 153.98 136.47 17.51 4 7 3 7 2 6
3 153.01 139.80 13.21 3 7 3 7 2 6
4 158.46 140.32 18.13 3 7 4 7 2 6

603
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

The hybrid model determined the best solution under stochastic factors minimizing the difference between the
deterministic and the stochastic model. For example, in the first scenario, the hybrid model reaches a decrease in the impact
of the stochastic factors as the candidate solutions are analyzed. Therefore, in iteration 3, the best solution is reached, since
the value of 𝑆𝑛 in this iteration is lower than the other three iterations, as shown in Figure 9.

Millons
160
140
120
Revenue

100
80
60
40
20
0
1 2 3 4
Iterations

MILP Simulation 𝑆𝑛
Sn

Figure 9. Impact of stochastic factors

The results obtained from the hybrid model for the first scenario determined that the distribution network should be
formed by 28 distribution centers to satisfy the demand of 27 markets and maximize the farmers’ profits depending on the
delivered quality, as shown in Figure 10.
The selected centers according to this model were: three in the State of Guanajuato, seven in Hidalgo, three in the State
México, seven in Puebla, two in San Luis Potosíand six in Zacatecas. The average use of the storage capacity of the selected
distribution centers was of 86% according to the simulation, as shown in Figure 11.
Concerning prickly pear quality, it is worth recalling that the quality was evaluated in a discrete and stochastic way
using Equation (5). For the distribution network design proposed in scenario 1, the simulation model presents the result of
the quality levels between 84% and 94% depending on the location of the market where the prickly pear is consumed, as
shown in Figure 12. For example, at market 22 the minimum quality the product can have considering the network design
suggested by the hybrid model is of 84.4% while the maximum quality that could be delivered to the market is of 92.5%. In
another case of this scenario, the quality levels that can be reached in the delivery happened at market 26 with a minimum
level of 88.5% and a maximum level of 94%.

Figure 10. The distribution network of scenario 1 (iteration 3)

604
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

100%

80%

60%

40%

20%

0%
Storage

Guanajuato Hidalgo Mexico Puebla San Luis Potosi Zacatecas

Figure 11. Utilization of distribution centers capacity

95

93

91
% Quality

Min
89
Ma x
87 Avera ge

85

83
1 3 5 7 9 11 13 15 17 19 21 23 25 27

Market

Figure 12. Prickly pear quality at the market

The other two scenarios were solved considering the same factors of costs and price but with similar variations of 20%
and 30%, respectively. In scenarios two and three, the results indicated that to maximize the profit of the distribution network
of prickly pears 29 and 28 distribution centers must be open, respectively. In scenario two, average quality levels of 87% and
90% were obtained, and for scenario three, levels between 86% and 89%. In scenario two, the minimum quality levels
happened at market 11. In this market, prickly pear arrives with quality of 85.7% and can reach a maximum quality of 92.3%.
On the other hand, market 18 presents the best quality levels with a minimum value of 88.1% and a maximum level of 94.4%.
Finally, in scenario three market 20 reached a minimum quality of 85.7% and a maximum quality of 91%, while the best
levels of quality occurred at market 17 with values of 88% and 95%. The results obtained for each scenario are shown in
Table 9.

605
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

Table 9. Results of the three scenarios

Scenario 1
Average
Millions Distribution centers
quality (%)
Iteration
MILP Simulation 𝑆𝑛 Min Max Guanajuato Hidalgo Mexico Puebla San Zacatecas
Luis
3 7 3 7 2 6
3 153.01 139.80 13.21 84 94 Capacity utilization (%)
88 79 76 91 92 93

Scenario 2
Millions Average Distribution centers
quality (%)
Iteration
MILP Simulation 𝑆𝑛 Min Max Guanajuato Hidalgo Mexico Puebla San Zacatecas
Luis
4 7 3 7 2 6
4 164.27 146.79 17.48 87 90 Capacity utilization (%)
86 84 78 98 80 93

Scenario 3
Millions Average Distribution centers
quality (%)
Iteration
MILP Simulation 𝑆𝑛 Min Max Guanajuato Hidalgo Mexico Puebla San Zacatecas
Luis
4 6 3 7 2 6
2 166.07 145.23 20.84 86 89 Capacity utilization (%)
79 80 76 99 87 98

The results previously described clearly show that prickly pear’s deterioration falls up to 84% because there is no
refrigeration in the storage and distribution, meaning that there is no cool chain, this has an impact on the selling price. This
situation is prevalent in developing countries as they are limited in infrastructure and technology because of the high costs of
the capital and operation of the refrigeration equipment that is needed in storage facilities and refrigerated freight vehicles.
In this case, producers, distributors, and retailers under these conditions must have strategies that let them deliver fresh food
of good quality to maximize their profits.
The results overall demonstrate the capability of the hybrid model to act as a decision making support tool for helping
management for efficiently finding a distribution network design to maximize stakeholders profits. The advantage of using
the hybrid model is that it allows incorporating stochastic factors that can affect the supply chain performance, one of the key
performance indicators of this type of chains is the quality of the food delivered, which is directly associated with the sale
price and with stakeholders profits.

5. CONCLUSIONS

In this paper, a hybrid model that integrates MILP and simulation is proposed to optimize the distribution network design of
the prickly pear supply chain. The main contribution of this paper to the literature is a model that evaluates food quality with
a discrete and stochastic approach along the whole supply chain with the objective maximizing the farmers' profits depending
on the quality delivered at the market. Also, the model that evaluates the prickly pear quality considering the two main
variables that affect the food freshness, which is temperature and time.
One of the main challenges for prickly pear supply chain in Mexico is that it does not consider refrigeration aspects in
the storage and transportation process. Also, farmers use one or more intermediaries to move their product causing longer
delivery times at the market. The hybrid model shows that the state of Zacatecas represents a pivotal location to ensure supply
in the central and northern regions of the country, since at this point at least 50% of the crop obtained from the different
producers selected in this study converge. In most cases, the utilization of distribution centers capacity in the model is covered
at 100%. However, the simulation shows that the prickly pear quality level can degrade up to 15% when it reaches the market.

606
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

Finally, we conclude that the proposed hybrid model is suitable in gaining insights in what the effect is of a food
distribution network design on its operational and economic performance of a supply chain that lacks cold chain
infrastructure. In this sense, the importance of storage and transport systems and conditions during the supply of food will
become even more critical as supply chains in markets become more coordinated and integrated. Therefore, it is necessary to
use analytical tools to manage the food distribution in supply chains where the cold chain is not used so that these chains
operate efficiently and offer quality food at competitive prices in a sustainable manner.
For future research, a different approach could be used in the optimization part. For example, stochastic programming could
be used to capture some of the stochastic factors before simulating a candidate solution. Also, some restrictions related to
inventory control, sustainability (energy consumption and GHG emissions), production planning and programming, vehicle
routing, facility location, delivery times, accepted quality level, among others, may be incorporated.

REFERENCES

Acar, Y., Kadipasaoglu, S.N., and Day, J.M. (2009). Incorporating uncertainty in optimal decision making: Integrating mixed
integer programming and simulation to solve combinatorial problems. Computers & Industrial Engineering, 56(1):106–112.

Accorsi, R., Cholette, S., Manzini, R., Pini, C., and Penazzi, S. (2015). The land-network problem: ecosystem carbon balance
in planning sustainable agro-food supply chains. Journal of Cleaner Production, 112:158–171.

Aghighi, A., and Malmir, B. (2016). Designing distribution networks of perishable products under stochastic demands and
routs. In Proceedings of the International Conference on Industrial Engineering and Operations Management (pp. 1008–
1019). Detroit, Michigan, USA.

Allaoui, H., Guo, Y., Choudhary, A., and Bloemhof, J. (2018). Sustainable agro-food supply chain design using two-stage
hybrid multi-objective decision-making approach. Computers and Operations Research, 89:369–384.

Almeder, C., Preusser, M., and Hartl, R.F. (2009). Simulation and optimization of supply chains: Alternative or
complementary approaches?. OR Spectrum, 31(1):95–119.

Ambekar, S., and Kapoor, R. (2019). Optimization of inventory policies of food grain distribution stage in public distribution
system. Benchmarking: An International Journal, 26(2): 692–713.

ASERCA (2017). Centros de Acopio. Retrieved 15 April, 2018, from http://www.aserca.gob.mx/comercializacion/acopio.

Ayllon, B.J.C., Omaña, S.J.M., Sangerman-Jarquín, D.M., Garza, B.L.E., Quintero, R.J.M., and González, R.F.J. (2015).
Modelo de transporte en México para la minimización de costos de distribución de tuna (Opuntia spp.) en fresco. Revista
Mexicana de Ciencias Agrícolas, 6(7):1615-1628.

Beshara, S., El-kilany, K.S., and Galal, N.M. (2012). Simulation of Agri-Food Supply Chains. World Academy of Science,
Engineering and Technology, 6(5):388–393.

CEC (2017). Characterization and Management of Food Loss and Waste in North America. Montreal, Canada: Commission
for Environmental Cooperation.

Chwif, L., Banks, J., Filho, J.P.M., and Santini, B. (2013). Framework for specifying a discrete-event simulation conceptual
model. Journal of Simulation, 7(1):50–60.

De Keizer, M., Akkerman, R., Grunow, M., Bloemhof, J.M., Haijema, R., and Van der Vorst, J.G.A.J. (2017). Logistics
network design for perishable products with heterogeneous quality delay. European Journal of Operational Research,
262(2):535-549.

Esteso, A., Alemany, M. M. E., and Ortiz, A. (2018). Conceptual framework for designing agri-food supply chains under
uncertainty by mathematical programming models. International Journal of Production Research, 56(13): 4418–4446.

607
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

FAO (2018). Pérdida y desperdicio de alimentos. Retrieved 15 April, 2018, from http://www.fao.org/food-loss-and-food-
waste/es/?cm_mc_uid=82170200740215056064001&cm_mc_sid_50200000=1505606400.

Ge, H., Nolan, J., Gray, R., Goetz, S., and Han, Y. (2016). Supply chain complexity and risk mitigation - A hybrid
optimization - simulation model. International Journal of Production Economics, 179:228–238.

González-Hernández, I. J., Martínez-Flores, J. L., Sánchez-Partida, D., and Gibaja-Romero, D. E. (2019). Relocation of the
distribution center of a motor oil producer reducing its storage capacity: A case study. Simulation, 1–16.

Granillo-Macias, R., Olivares-Benitez, E., Flores-Martinez, J.L., and Caballero-Morales, S.O. (2018). Analysis of logistic
cost in contract agriculture: the case of barley supply chain in Hidalgo, México. Custos e Agronegocio On-line, 14(1):164-
183.

Jahani, H., Abbasi, B., Alavifard, F., and Talluri, S. (2018). Supply chain network redesign with demand and price
uncertainty. International Journal of Production Economics, 205:287–312.

Jan, O., Tostivint, C., Turbé, A., O´Connor, C., and Lavelle, P. (2013). Food wastage footprint - Impacts on natural resources
- Summary Report. Food and Agriculture Organization of the United Nations: FAO.

Jonkman, J., Bloemhof, J. M., van der Vorst, J. G. A. J., and van der Padt, A. (2017). Selecting food process designs from a
supply chain perspective. Journal of Food Engineering, 195:52–60

Jonkman, J., Barbosa-Póvoa, A. P., and Bloemhof, J. M. (2019). Integrating harvesting decisions in the design of agro-food
supply chains. European Journal of Operational Research, 276(1): 247–258.

Ling, B., Tang, J., Kong, F., Mitcham, E.J., and Wang, S. (2015). Kinetics of Food Quality Changes During Thermal
Processing: a Review. Food and Bioprocess Technology, 8(2): 343–358.

Manzini, R., Cholette, S., Accorsi, R., Penazzi, S., and Pini, C. (2015). The land-network problem: ecosystem carbon balance
in planning sustainable agro-food supply chains. Journal of Cleaner Production, 112:158–171.

México Produce (2016). El nopal y tuna. Retrieved 28 May, 2018, from http://www.mexicoproduce.mx/2016/10/el-nopal-y-
tuna.html.

Micale, R., and La Scalia, G. (2018). Shelf life-based inventory management policy for RF monitored warehouse.
International Journal of RF Technologies-Research and Applications, 9(3–4): 101–111.

Miranda-Ackerman, M. A., Fernández-Lambert, G., Azzaro-Pantel, C., and Aguilar-Lasserre, A.A. (2014). A Multi-
Objective Modelling and Optimization Framework for Operations Management of a Fresh Fruit Supply Chain: A Case Study
on a Mexican Lime Company. Applications of Metaheuristics in Process Engineering, Valadi, J. & Siarry, P. (eds), Springer.

Miranda-Ackerman, M. A., Azzaro-Pantel, C., and Aguilar-Lasserre, A. A. (2017). A green supply chain network design
framework for the processed food industry: Application to the orange juice agrofood cluster. Computers and Industrial
Engineering, 109:369–389.

Nguyen, C., Dessouky, M., and Toriello, A. (2014). Consolidation strategies for the delivery of perishable products.
Transportation Research Part E: Logistics and Transportation Review, 69:108–121.

Noordhoek, M., Dullaert, W., Lai, D. S. W., and de Leeuw, S. (2018). A simulation–optimization approach for a service-
constrained multi-echelon distribution network. Transportation Research Part E: Logistics and Transportation Review,
114:292–311.

Nunes, M.C.D.C., Nicometo, M., Edmon, J.P., Melis, R.B., and Uysal, I. (2014). Improvement in fresh fruit and vegetable
logistics quality: berry logistics field studies. Philosophical Transactions of the Royal Society A: Mathematical, Physical and
Engineering Sciences, 372(2022):1–19.
608
González-Hernández et al. Hybrid Model to Design An Agro-Food Distribution Network

Petig, E., Rudi, A., Angenendt, E., Schultmann, F., and Bahrs, E. (2018). Linking a farm model and a location optimization
model for evaluating energetic and material straw valorization pathways—A case study in Baden-Wuerttemberg. GCB
Bioenergy, 11(1): 304–325.

Pinedo-Espinoza, J.M., Franco-Bañuelos, A., and Hernández-Fuentes, A.D. (2010). Comportamiento poscosecha de
cultivares de tuna por efecto del manejo del huerto y temperatura de frigo conservación. Revista Iberoamericana de
Tecnología Postcosecha, 11(1): 43-58.

Pourya, P., and Kyoung, K. (2016). The New Generation of Operations Research Methods in Supply Chain Optimization: A
Review. Sustainability, 8:1-23.

Ramírez-Abarca, O., Figueroa-Hernández, E., and Espinosa-Torres, L.E. (2015). Análisis de rentabilidad de la tuna en los
municipios de Nopaltepec y Axapusco, Estado de México. Revista Mexicana de Agronegocios, XIX(36):1199-1210.

Sazvar, Z., Rahmani, M., and Govindan, K. (2018). A sustainable supply chain for organic, conventional agro-food products:
The role of demand substitution, climate change and public health. Journal of Cleaner Production, 194:564–583.

Sel, Ç ., Bilgen, B., and Bloemhof-Ruwaard, J. (2017). Planning and scheduling of the make-and-pack dairy production under
lifetime uncertainty. Applied Mathematical Modelling, 51:129–144.

SIAP (2018). Servicio de Información Agroalimentaria y Pesquera, SAGARPA. Retrieved 7 January, 2018, from
http://www.sagarpa.gob.mc/quienesomos/datosabiertos/siap/Paginas/estadistica.aspx.

Soto-Silva, W. E., Nadal-Roig, E., González-Araya, M. C., and Pla-Aragones, L. M. (2016). Operational research models
applied to the fresh fruit supply chain. European Journal of Operational Research, 251(2): 345–355.

Sutopo, W., Hisjam, M., and Yuniaristanto. (2013). An Agri-food Supply Chain Model to Empower Farmers for Supplying
Deteriorated Product to Modern Retailer. IAENG Transactions on Engineering Technologies, Yang, G-C., Ao, S-I., Huang,
X. & Castillo, O. (eds), Lecture Notes in Electrical Engineering, Springer.

Tirkolaee, E. B., Goli, A., Bakhsi, M., and Mahdavi, I. (2017). A robust multi-trip vehicle routing problem of perishable
products with intermediate depots and time windows. Numerical Algebra, Control and Optimization, 7(4): 417–433.

Tsao, Y-C. (2013). Designing a Fresh Food Supply Chain Network: An Application of Nonlinear Programming. Journal of
Applied Mathematics, 2013:1–8.

Van Boekel, M.A.J.S. (2008). Kinetic Modeling of Food Quality: A Critical Review. Comprehensive Reviews in Food Science
and Food Safety, 7(1):144–158.

Wang, X., Wang, M., Ruan, J., and Zhan, H. (2016). The Multi-objective Optimization for Perishable Food Distribution
Route Considering Temporal-spatial Distance. Procedia Computer Science, 96:1211–1220.

Zhao, R., Liu, Y., Zhang, Z., Guo, S., Tseng, M. L., and Wu, K. J. (2018). Enhancing eco-efficiency of agro-products’ closed-
loop supply chain under the belt and road initiatives: A system dynamics approach. Sustainability, 10(3): 1–15.

Zhu, L., and Lee, C. (2018). Analysis of a traceability system for perishable food supply chains. International Journal of
Industrial Engineering: Theory, Applications and Practice, 25(1):54-66.

609

You might also like