Sage Journals: Discover world-class research

Abstract

Properly locating these facilities is a substantial factor in the success of the logistics systems. In this paper, a bi-objective mathematical model for a maximal covering hub location problem is presented to minimize time and environmental risks. The Goal Attainment method was employed to solve the small-sized problems for model validation. Since the problem is NP-Hard, the Multi-Objective Imperialist Competitive Algorithm (MOICA) meta-heuristic algorithm was exploited for solving the medium and large-sized problems. The performance of MOICA was compared with the performance of the Goal Attainment method and the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm to validate the proposed model and solution approach. This paper can direct the logistics companies to reduce the cost, time, and environmental effects of their transportation networks. In addition, this research can optimize energy consumption in the transportation sector for the continuation of low-cost services and reduce fuel consumption, which leads to reducing environmental pollution.

Keywords

hub covering location problem bi-objective mathematical programming model goal attainment method;meta-heuristic algorithm multi-objective imperialist competitive algorithm multi-objective particle swarm optimization

Introduction

The problem of locating facilities is one of the issues that should be considered in the initial stages of designing industrial systems. Choosing an appropriate location for a facilitation center can be an important factor in its success. Also, the hub location problem (HLP) is one of the issues raised in the field of location (Yin et al., 2019; Soleimani et al., 2022). This topic is very important because of its broad applications. Locating hubs in transportation systems has been one of the most challenging and significant issues in recent years (K. Yang et al., 2014). Hubs are facilities that serve as transfer and exchange points to multiple origins and destinations for remote transportation and communication systems (Yaman & Elloumi, 2012). The main advantage of using the hub network is its economic efficiency due to the indirect connection between the origin and destination points. In the hub network, it is usually assumed that the use of inter-hub transportation leads to cost reduction, which is economical compared to direct transmission networks. In transportation networks, where there is a large volume of transportation between several points, a hub can be established for economic reasons to decrease transportation time, cost, and fuel consumption (J. Wang & Qin, 2021; Zahedi & Ghahremani Nahr, 2020).

The HLP focuses on how to establish hubs to optimize the entire system. In such systems, direct transportation flow does not exist between two nodes (origin and destination), but through one or two hubs (Qu & Weng, 2009). In the HLP, the service includes the movement of people, goods, or information between the place of origin and destination (Vural & Aygün, 2019). Hub equipment is used as distribution and collection centers in many transportation networks (Tofighian & Arshadi khamseh, 2020). These types of systems are addressed when goods, passengers, or information must be sent from a set of origins to a set of destinations (Tikani et al., 2021). Facilities that perform the operations of exchange, handling, collection, classification, loading, and distribution are designed as hubs (Talbi & Todosijević, 2017). In fact, hubs are exploited as the only valid intermediate points on the route from origin to destination. Flows from different origins are first collected in these hubs to pass through an intermediate hub or to be sent to the final destination, and they are combined and sent with flows that have the same destination and arrived from other origins (Rostami et al., 2021). Usually, the main goal of hub network design is locating hub facilities and assigning non-hub nodes to hubs in building the network with the lowest cost in accordance with the cost of deployment and transportation (de Camargo et al., 2009).

In most of the manufacturing factories and service companies, timely supply of customer orders or timely service is important. Costs of delay in these problems not only harm the customer but also reduce the reputation of service companies or manufacturing factories. Transportation scheduling problems seek to find optimal scheduling with the aim of reducing transportation times and costs (Grine et al., 2021).

In addition, due to the importance of reducing environmental pollution and environmental and social pressures, there is growing attention to environmental issues (Eydi & Nasiri, 2019). Transportation with heavy machinery can have harmful effects on the environment, such as noise pollution and air pollution caused by the emission of greenhouse gases like $C O_{2}$ (Amin-Naseri et al., 2018). Therefore, optimizing energy consumption in transportation systems to reduce production cost and fuel consumption, which leads to decreasing environmental pollution, is essential for manufacturing industries (Akgün & Tansel, 2018; Brimberg et al., 2022).

Therefore, a mathematical programming model is presented in this study for the HLP under uncertainty conditions to reduce time, cost, and environmental impact of the transportation network. In the present research, a central (focal) factory together with a number of hubs (for example, a subsidiary factory) connecting to a number of nodes (for example, customers, or stores) are considered in a network. The central factory is in charge of processing raw materials. Then, these materials are transferred to the factories where they are transformed into finished products. Finally, finished products are transferred from the hubs to the stores or customers (nodes).

The problem investigated in this study is a single-allocation, bi-objective coverage HLP with capacity constraints. The first objective function of the proposed model is to reduce costs. Due to the fact that only few criteria have been considered for the design of hub networks so far, the second objective function of this model seeks to reduce environmental impact. Also, the second objective function attempts to select the hub nodes that are in a better condition based on the four criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability. Due to the computational complexity of the proposed model, the well-known meta-heuristic algorithm MOICA is exploited to find near-optimal solutions. Finally, the solutions obtained by the meta-heuristic algorithm are compared with the solutions obtained by the exact method for small and medium-sized problems.

The objectives of this research can be stated as follows:

- Designing an efficient hub network for the transfer of raw materials from the focal factory to the hubs,

- Minimizing the costs of the transportation network,

- Minimizing the costs and time of transferring raw materials from the focal factory to stores or customers,

- Reducing the environmental impact of the transportation network,

- Increasing the service quality of the transportation network,

- Reducing the traffic of the transportation network,

- Increasing the utilization of the transport infrastructure of the transport network,

- Increasing the expandability of the transportation network and,

- Considering the conditions of uncertainty to deal with real world changes.

The most important questions of the current study can be mentioned as follows:

- What effect does the use of hubs have on the cost and time of material transportation?

- What is the effect of the uncertainty of the model parameters on the network structure?

- How to assess the environmental hazards in the HLP?

- hat are the effects of the four criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability on the transportation network?

The main contributions of this research can be expressed as follows:

- Presenting an optimization model for the HLP under uncertain and fuzzy environment,

- Using the four criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability to calculate the weights of different nodes and using these weights in a separate objective function of the bi-objective coverage HLP.

This paper is structured as follows. In the following section, the relevant studies are reviewed. In Section “Problem definition,” the problem definition and mathematical programming model is given for the problem in hand. Section “Solution approach” explains the solution approach. In Section “Results,” the numerical examples are provided for the validation of the proposed model and solution method, also, the results are illustrated. Subsequently, a real case study is provided in Section “Case study.” Finally, Section “Conclusion” concludes the paper.

Literature Review

Reviewing the Hub Location Problems (HLPs)

In a large number of network design problems such as location, transportation, production planning and scheduling, finance, engineering design, etc., uncertainty should be considered to increase the efficiency of the system during operations. For this purpose, many researchers have addressed uncertainty in the HLPs. In this section, the relevant studies on the HLPs are reviewed.

T. H. Yang (2009) investigated the hub problem in air transportation. He considered the probabilistic demand with seasonal changes. The two-stage probabilistic model was s solved considering three scenarios. Also, Sim et al. (2009) developed a model for the p-hub center problem considering normal probability distribution for travel times and service limitation. In addition, Contreras et al. (2011) studied the HLP with multiple allocations and probabilistic data without considering facility capacity constraints. In this study, uncertainty in demand and transportation costs was taken into account. Moreover, Alumur et al. (2012) considered single allocation and multiple allocation HLPs separately without capacity constraints under uncertainty in the cost of preparing facilities and demand. Furthermore, K. Yang et al. (2013) presented a fuzzy model for the p-hub center problem considering perishable goods. The travel times were considered non-deterministic in the form of fuzzy numbers. They solved the model with a combination of Particle Swarm Optimization (PSO) algorithm with Genetic operators and local search.

Several researchers and scholars tackled the uncertainty in the HLP with three different modeling approaches including Stochastic Programming, Fuzzy Programming, and Robust Optimization (Abbasi-Parizi et al., 2018; Li et al. (2020); Maharjan & Hanaoka, 2020; Nourzadeh et al. (2020); Rahmati & Neghabi, 2021; Rajabi & Avakh Darestani, 2020; Shang et al., 2021a, 2021b; Snyder, 2006; Vahdani & Mohammadi, 2015; J. Wang & Qin, 2021; Zahedi & Ghahremani Nahr, 2020).

Rahmanniyay et al. (2020) presented a multi-objective integer linear programming model for a multi-valued p-hub median problem under uncertainty. In this model, the uncertainty in demand, transportation time, and establishment cost was considered. This model aimed to minimize the cost of transportation and establishment as well as to minimize the delay time of the distribution network in the supply chain. Mumtaz and Eriskin (2021) introduced a nonlinear optimization model to minimize the total population covered considering the service requirement. Alumur et al. (2021) provided a roadmap and made some suggestions for further studies on the HLPs. Shang, Jia, et al. (2021) proposed a credit-based fuzzy programming model with the objective of minimizing the last arrival time of cargo for the hierarchical multimodal HLP under uncertainty in cargo delivery systems. They considered the uncertainty of travel time and displacement time as fuzzy numbers and employed a heuristic method to solve the problem. Chobar et al. (2022) developed a bi-objective mathematical programming model to decrease the network cost and greenhouse gas emissions in the hub-spoke network for the perishable tourism products. Also, Ghaffarinasab et al. (2023) introduced the multi-allocation p-hub maximal covering problem in which the traffic of each origin–destination pair is classified into different groups with specified service level requirements. Moreover, Maleki et al. (2023) addressed the multiple allocation maximal hub covering problem where the flow of each origin–destination pair is distributed through various hubs in a way that the total flow covered by the located hubs is maximized. Furthermore, H. Wang et al. (2024) proposed a Lattice Boltzmann Method-like algorithm for the maximal covering location problem.

Reviewing the Solution Approaches

Except for some special cases, several authors proved that the HLP is NP-hard (Kara & Tansel, 2003; Weng et al., 2006). For this purpose, researchers have used different solution methods to solve these problems. In the following, some of these different solution methods used in the literature are reviewed.

Heuristic algorithms seek to find the optimal solution in the search space by putting together a series of rules and using a series of additional information to guide the search. Although these algorithms do not guarantee finding the optimal solution, they can find a good solution in a reasonable time. During the last three decades, new optimization methods have been introduced that seek to combine basic heuristic methods at a higher level to make the search more efficient, and these methods have become known as meta-heuristic algorithms. Meta-heuristic algorithms are general methods that can be used to solve various problems without the need for fundamental changes in their framework.

It has been proven that the p-hub median problem is also NP-hard. In this problem, even if the locations of the hubs are known in advance, the allocation problem is still NP-hard (Kara & Tansel, 2000, 2003). O’Kelly (1987) introduced two heuristic algorithms for the p-hub median problem with single assignment. Subsequently, Klincewicz (1991, 1992) proposed several heuristic algorithms for the p-hub median problem with single assignment. Also, Skorin-Kapov et al. (1996) used a Tabu Search algorithm for solving the p-hub median problem with single assignment. In addition, Abdinnour-Helm and Venkataramanan (1998) employed Simulated Annealing (SA) algorithm for solving the p-hub median problem with single assignment. Moreover, Sasaki et al. (1999) presented a greedy algorithm for the p-hub median problem with multiple assignment. Furthermore, Abdinnour-Helm (2001) suggested a heuristic algorithm based on the combination of Genetic Algorithm (GA) and Tabu Search (TS) algorithm to solve the HLP considering the fixed costs.

The first heuristic algorithm for the p-hub center problem with single assignment was proposed by Pamuk and Sepil (2001). The algorithm was a single-relocation algorithm for making decisions about location and allocation in acceptable time. They combined the algorithm with TS to increase the possibility of avoiding local optimal solutions. Then, Topcuoglu et al. (2005) exploited the Genetic Algorithm (GA) for the single assignment HLP considering the fixed costs without capacity constraints. The GA algorithm obtained much better solutions with less processing time compared to the hybrid algorithm introduced by Abdinnour-Helm and Venkataramanan (1998). Subsequently, Cunha and Silva (2007) used a Genetic Algorithm (GA) combined with Simulated Annealing (SA) algorithm to solve this problem. Also, another algorithm to solve this problem was introduced by Chen (2007), which was a combination of Tabu Search (TS) algorithm, Simulated Annealing (SA) algorithm, and advanced methods. This algorithm had advantages in terms of solution quality and solution time compared to the algorithm presented by Topcuoglu et al. (2005). Due to the fact that no comparison has been made with the algorithms introduced by Cunha and Silva (2007) and Chen (2007), it can be concluded that these two algorithms are the best algorithms presented so far to solve the single assignment HLP with fixed costs. Calık et al. (2009) utilized a TS-based heuristic algorithm to tackle this problem. Also, Gavriliouk (2009) used another heuristic algorithm for dealing with the p-hub center problem. Ilić et al. (2010) employed a Variable Neighborhood Search (VNS) method for the p-hub median problem with single assignment. Then, Parvaresh et al. (2014) studied the bi-level HLP and solved it by using the SA and TS algorithms.

Problem Definition

In this study, a central factory and a number of hubs (for example, affiliated or subsidiary factory) that are connected to a number of nodes (for example, a store or a customer) in a network are considered. The central factory is in charge of processing raw materials. Then, raw materials are transferred to affiliated factories where they are transformed into finished products. Finally, the finished products are transferred from the affiliated factories (hubs) to the stores or customers (nodes) with the least cost and time. As an instance, Figure 1 shows the location of the central factory and the nodes. Considering the location of the central factory and the cost of transporting raw materials from the central factory to each of the nodes, the hub network is designed for the existing nodes and the hub nodes are identified. Figure 2 depicts the location of the central factory and the hub network.

Figure 1.

Location of the central factory and nodes.

Figure 2.

Location of the central factory and the designed hub network.

As shown in Figure 2, a single-assignment bi-objective coverage HLP with capacity limitation is investigated. The first objective function of the proposed model is to minimize costs. Due to the fact that only a few criteria have been considered for the design of hub networks so far, the second objective function of the model is to minimize the environmental impacts caused by the construction of hubs (due to locating the production facilities) and seeks to consider hub nodes that are in a better condition based on four criteria of service quality, traffic in the covered area, transportation infrastructure, and expandability. In addition, in the studied HLP, the hub nodes are fully connected to each other and each non-hub node is assigned to exactly one hub. In this problem, a constant cost reduction coefficient $α \in [0, 1]$ is used for hub-to-hub connections and a constant cost reduction coefficient $γ \in [0, 1]$ is used for non-hub to hub connections, so that $γ > α$ . Also, direct connection between non-hub nodes is not allowed, and for each hub, a radius to cover non-hub nodes (covering radius) and a capacity limit are considered. Therefore, non-hub nodes can be assigned to a hub only when they are within the coverage area of that hub and that hub has the necessary capacity. Also, no structure has been imposed on the hub network and the number of hubs is not predetermined. As shown in Figure 3, each dashed line represents a direct route between the central factory and the hubs (affiliated factories).

Figure 3.

Direct routes between the central factory and the hubs.

Proposed Mathematical Programming Model

In this section, the mathematical programming model is presented for the bi-objective coverage HLP with single assignment and capacity limitation. The first objective is to reduce costs. Due to the fact that so far only a few criteria have been considered for the design of hub networks, the second objective of the model seeks to reduce environmental risks and to make hub nodes that are in a better condition in terms of four quality criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability.

First, the assumptions of the model are provided. Then, the indexes, parameters, and decision variables of the model are defined. Finally, the model for the problem in hand is proposed.

Assumptions

- The location (position) of the points is definite and deterministic, also, any point can be selected as a hub.

- The number of hubs is not known in advance.

- At least one hub must be built.

- Hub nodes are fully connected to each other (full communication).

- Direct communication between non-hub nodes is not allowed and transfer is done only through intermediate nodes (hubs).

- Allocation of non-hub nodes to hubs is single, which means that each non-hub node can transfer goods to only one hub and receive goods from only one hub (single allocation).

- Each hub can receive a certain amount of incoming flow according to its capacity limitation.

- Only non-hub nodes that are within the coverage radius of a hub can be assigned to the desired hub (covering hub).

- There is a fee for establishing hubs in the network.

- A reduction discount coefficient of $α$ is given for the flow exchange between two hubs.

- The reducing discount rate $γ$ is assigned to the hub for the flow exchange between non-hub nodes.

- Production facilities (affiliated or subsidiary factories) are located in hubs to reduce costs.

- The risk assessment criterion is the fuzzy risk priority number for the environmental risks caused by the establishment of the hub in each node due to the location of production facilities (affiliated factories) in that hub.

- The weight of the hub deployment in each node is considered according to four criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability.

- A fee is considered for the transfer of raw materials from the central factory to each of the hubs (affiliated or subsidiary factories).

- No structure is imposed on the hub network.

The definition of the indexes, parameters, and decision variables of the model are provided as follows.

Indexes
$i, j \in N = {1, 2, \dots, n}$ :	Set of non-hub nodes
$k, l \in N = {1, 2, \dots, n}$ :	Set of non-hub nodes
Parameters
$V_{ij}$ :	The amount of flow between node i and node j
$C_{ij}$ :	The cost of flow between two nodes i and j per unit of flow
$F_{k} :$	The fixed cost of establishing the hub in node k
$r_{k} :$	The highest collection/distribution cost between hub k and the nodes assigned to hub k (covering radius of hub k)
$Chu b_{k} :$	Capacity of hub k
$O_{i} = \sum_{j \in N} V_{ij} :$	The sum of flows originating from node i
$α :$	Cost discount coefficient between two hubs
$γ :$	Cost discount coefficient between non-hub and hub nodes
$b_{k} :$	The cost of transferring any type of raw material from the central factory to the hub k
$M :$	A very large number
$P :$	All kinds of raw materials
$\tilde{RP N_{k}} :$	Fuzzy risk priority number related to the environmental risks caused by the establishment of the hub in the node k
$W_{k} :$	Hub establishment weight in node k according to the four quality criteria
Decision variables
$X_{ik} :$	Equal to one if node i is assigned to hub k, otherwise equal to zero
$X_{kk} :$	If node k becomes a hub, it is equal to one, otherwise it is equal to zero
$Y_{kl}^{i} :$	The amount of flow originating from node i and transported through hubs k and l

Mathematical Programming Model

\begin{matrix} Min Z_{1} = \sum_{i = 1}^{n} \sum_{k = 1}^{n} γ C_{ik} X_{ik} O_{i} + \sum_{i = 1}^{n} \sum_{k = 1}^{n} \sum_{l = 1}^{n} α C_{kl} Y_{kl}^{i} \\ + \sum_{k = 1}^{n} F_{k} X_{kk} + \sum_{k = 1}^{n} P O_{k} b_{k} X_{kk} \end{matrix}

(1)

Min Z_{2} = \sum_{k = 1}^{n} [(\frac{{\tilde{RPN}}_{k}}{\sum_{k = 1}^{n} {\tilde{RPN}}_{k}}) + (1 - W_{k})] X_{kk}

(2)

s.t.

\sum_{k = 1}^{n} X_{ik} = 1 \forall i

(3)

\sum_{k = 1}^{n} X_{kk} \geq 1

(4)

\sum_{i = 1}^{n} X_{ik} \geq 1 \forall k

(5)

\sum_{l = 1}^{n} Y_{kl}^{i} + \sum_{j = 1}^{n} V_{ij} X_{jk} = O_{i} X_{ik} + \sum_{l = 1}^{n} Y_{lk}^{i} \forall i, k

(6)

\sum_{\begin{matrix} i = 1 \\ i \neq k \end{matrix}}^{n} O_{i} X_{ik} \leq Chu b_{k} \forall k

(7)

r_{k} \geq C_{ik} X_{ik} \forall i, k

(8)

X_{ik} \leq X_{kk} \forall i, k

(9)

Y_{kl}^{i} \leq M X_{kk} \forall i, k, l

(10)

Y_{kl}^{i} \leq M X_{ll} \forall i, k, l

(11)

Y_{kl}^{i} \leq M X_{ik} \forall i, k, l

(12)

X_{ik} \in {0, 1} \forall i, k

(13)

Y_{kl}^{i} \geq 0 \forall i, k, l

(14)

Equation 1 states the first objective function of the model that minimizes the total costs in which the first sentence calculates the whole costs of flow among non-hub nodes and hub nodes. In the second sentence, the total costs of flow between the hub nodes are computed. In the third sentence, the total costs of establishing the hubs is measured, and in the fourth sentence, the total transportation costs of raw materials from the focal factory to the hubs (affiliated or subsidiary factories) is calculated.

Equation 2 shows the second objective function of the model that minimizes the environmental risks in which the first sentence calculates the total Fuzzy Risk Priority Number (FRPN) related to the environmental risks caused by establishing hubs. (It should be noted that FRPN is obtained by using the extended Fuzzy Failure Mode and Effects Analysis (FMEA) technique which was developed by Fattahi and Khalilzadeh (2018). For further details about the calculation method of FRPN, we refer to Fattahi and Khalilzadeh (2018)). In the second sentence, the total weight of the establishment of hubs is computed according to the four criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability.

It should be noted that the two objective functions are in conflict with each other, in that the nodes that have a higher Risk Priority Number (RPN), the cost of building hub in those nodes is lower and the nodes that have a lower Risk Priority Number (RPN), the cost of building hub in those nodes is higher. Also, the nodes in which the weight of building hub is higher according to the aforementioned four quality criteria, the cost of building hub in those nodes is also higher, but due to the fact that in the second sentence of the objective function, instead of $W_{k}$ , $(1 - W_{k})$ is used, as a result, the increase in the weight of building hubs in the second objective function reduces the value of the second objective function, while at the same time the cost of building hubs in the first objective function increases and reducing the weight of building hubs in the second objective function increases the value of the second objective function, while at the same time the cost of building hubs in the first objective function decreases.

Constraints (3) and (4) guarantee that each node is assigned to exactly one hub (single assignment). Constraint (5) indicates that at least one hub should be built in the network. Constraint (6) guarantees that at least one node is connected to each hub. Constraint (7) shows the flow balance. Constraint (8) states the capacity limitation of each hub. Constraint (9) dictates that node i can be assigned to hub k only when the transportation cost of a unit of flow between two nodes i and k ( $C_{ik})$ is equal to at most the covering radius of the hub k $(r_{k})$ . Constraint (10) shows that a node can be assigned to another node only when that node is a hub. Also, constraints (11), (12), and (13) mandate that the variable $Y_{kl}^{i}$ can have a positive value only when k and l are hubs. Finally, the constraint (14) states that the variable $Y_{kl}^{i}$ is positive, because this variable represents the flow value and the flow value must be positive (greater than zero).

In the above mathematical programming model, the improved Fuzzy FMEA method introduced by Fattahi and Khalilzadeh (2018) was used to calculate the Fuzzy Risk Priority Numbers (FRPNs) related to the total environmental risks caused by building hubs. Also extended Fuzzy MULTIMOORA and the extended Fuzzy Analytic Hierarchy Process (FAHP) methods introduced by Fattahi and Khalilzadeh (2018) as two extended fuzzy multi-criteria decision-making methods were applied to calculate the weights of building hubs in the nodes based on the four criteria of service quality, traffic of the covered area, transportation infrastructure, and expandability.

Solution Approach

Also, an exact Goal Attainment method or Chebyshev Goal Programming (Min-Max Goal Programming) is utilized to solve the small-sized problems. Due to the NP-hardness of the problem in hand, the MOICA and MOPSO meta-heuristic algorithms are exploited to solve the problem.

Goal Attainment or Chebyshev Goal Programming Method

Goal Programming is one of the most important multi-objective models. This method shows the way to move towards several conflicting and multiple objectives at the same time. Unlike linear programming that maximizes or minimizes the objective function value, Goal Programming minimizes the deviations between desired goals and actual results. The basis of the Goal Programming method is that it considers numbers as expectation levels for the objective functions of the problem. Then, it seeks to minimize the sum of deviations from these levels. One type of Goal Programming methods is the Goal Attainment or Chebyshev Goal Programming method. This method attempts to minimize the maximum deviations of each objective function. For this reason, this method is also called Min-Max Goal Programming. The general form of this method is as follows:

Min D

(15)

s.t.

h_{k} (x) = (\leq or \geq) 0 k = 1, 2, \dots, q

(16)

f_{i} (x) - d_{i}^{+} + d_{i}^{-} = b_{i} i = 1, 2, \dots, n

(17)

D \geq \sum_{i = 1}^{n} W_{i} (d_{i}^{+} + d_{i}^{-})

(18)

d_{i}^{+} + d_{i}^{-} \geq 0 i = 1, 2, \dots, n

(19)

In the above model, $D$ is a continuous variable that measures the amount of maximum deviations. In addition, $h_{k} (x)$ is the $K$ ^th system constraint and $f_{i} (x)$ is the i^th objective. $b_{i}$ is the expectation level of the i^th objective and $d_{i}^{+}$ are positive deviations and $d_{i}^{-}$ are negative deviations from the value of the i^th expectation level (goal), which are calculated as follows:

d_{i}^{-} = {\begin{matrix} b_{i} - f_{i} (x) if f_{i} (x) < b_{i} \\ 0 otherwise \end{matrix}

(20)

d_{i}^{+} = {\begin{matrix} f_{i} (x) - b_{i} if f_{i} (x) > b_{i} \\ 0 otherwise \end{matrix}

(21)

Also, $W_{i}$ are positive weight coefficients that determine the importance of each goal compared to other goals. It should be noted that in the proposed exact method, Pareto solutions are obtained by changing these weight coefficients and creating a loop with coding in MATLAB software.

The Multi-Objective Imperialist Competitive Algorithm (MOICA) Meta-Heuristic Algorithm

Imperialist Competitive Algorithm (ICA), like other evolutionary algorithms, starts with an initial population, each member of the population is called a country, and each empire consists of a number of colonies and an imperialist (colonizer) (Kaushal & Devi, 2023). The details and explanations of the ICA algorithm can be found in K. Yang et al. (2013) and Hosseini & Al Khaled (2014).

The Multi-Objective Particle Swarm Optimization (MOPSO) Algorithm

In the Particle Swarm Optimization (PSO) algorithm, each member of the population is called a particle. Each particle has three properties: position, velocity, and best experienced position. The particle calculates the value of the objective function in the position of the space where it is located, and since the particles are connected, the particle is informed of the best swarm position. Then, using the combination of the present information and the best location it has been in before, as well as the information of the best particles in the swarm, it chooses a direction to move. After performing the collective movement, one step of the algorithm ends. These steps are repeated several times until the desired solution is obtained. The details and explanations of the PSO algorithm can be found in Kennedy and Eberhart (1995) and Tang et al. (2016).

Solution Representation Method

It is obvious that the efficiency of an algorithm and the quality of the solutions are completely dependent on the solution representation method. Also, the solution representation should be such that the algorithm can search the solution space easily. The solution representation in the optimization problems are categorized into two classifications: discrete and continuous solution representation. In this research, the continuous solution representation method is used for both MOPSO and MOICA meta-heuristic algorithms. The solution representation method used for these two algorithms is as follows:

First, an $n \times n$ matrix is formed with random numbers between zero and one, where n represents the number of nodes. Table 1 displays a sample matrix with 5 nodes.

Table 1.

A Sample Matrix for Displaying a Solution in the HLP.

0.345	0.532	0.126	0.396	0.032
0.288	0.523	0.297	0.111	0.547
0.761	0.249	0.067	0.732	0.429
0.959	0.811	0.873	0.451	0.66
0.775	0.599	0.023	0.139	0.742

In the above matrix, each column represents one of the nodes. In other words, the column 1 displays node 1, the column 2 represents node 2, and so on. In the next step, the numbers on the main diameter of this matrix which are greater than 0.5, become 1, and the rest of the numbers on the main diameter (which are less than 0.5) become zero. The nodes whose numbers on the main diameter are 1 are hubs and the rest of the nodes are non-hub nodes. It should be noted that if none of the numbers on the main diameter are greater than 0.5, the largest number is selected and converted to 1, because there must be at least one hub. In this sample matrix, nodes 2 and 5 are hubs and nodes 1, 3 and 4 are non-hub nodes. Table 2 shows the above operation.

Table 2.

Identification of Hubs and Non-Hub Nodes.

0	0.532	0.126	0.396	0.032
0.288	1	0.297	0.111	0.547
0.761	0.249	0	0.732	0.429
0.959	0.811	0.873	0	0.66
0.775	0.599	0.023	0.139	1

In the next step, in order to determine which hub serves each node, the largest number in the column corresponding to that node (only one number) should be selected. If this maximum number is on the row of each hub, that node is assigned to the hub. As shown in Table 3, nodes 1 and 4 are assigned to hub 5 and node 3 is assigned to hub 2.

Table 3.

Assigning Non-Hub Nodes to Hubs.

0	0.532	0.126	0.396	0.032
0.288	1	0.297	0.111	0.547
0.761	0.249	0	0.732	0.429
0.959	0.811	0.873	0	0.66
0.775	0.599	0.023	0.139	1

After solution representation, the MOPSO and MOICA meta-heuristic algorithms go through their steps to find Pareto optimal or near optimal solutions.

Setting Parameters by the Taguchi Method

Design of experiments is one of the common techniques for identifying effective and ineffective factors, controlling effective factors and reducing ineffective factors to optimize the process. The purpose of designing experiments is to determine the variables that have the greatest effect on the response. In the design of the experiments, a series of tests are performed in which the desired changes are applied to the input variables in such a way that we can observe and determine the causes of the changes in the response. One of the response optimization techniques is the Taguchi method. The purpose of this method is to select the best combination of the parameters based on the execution of a number of experiments. In this method, the value of each parameter can be considered at several levels, for example, low, medium, and high. In each experiment, the value of the response variable should be converted according to the Taguchi method to the ratio of signal (S) to noise (N) and analyzed according to its changes. S means desirable values (response variable) and N means undesirable values (standard deviation). As a result, the goal is to maximize the signal-to-noise ratio. These signal-to-noise ratios are derived from the quadratic loss function using Equation 22:

SN = - 10 \log (\frac{1}{n} \sum_{i = 1}^{n} y_{i}^{2})

(22)

The analysis of the Taguchi method is based on two graphs, the first graph shows the main effects of the average solutions obtained, and the second graph shows the relationship between the levels of parameter values on the Signal-to-Noise ratio. In most cases, the results of the two graphs are the same, and if there are discrepancies, it is necessary to conduct new tests to make a decision.

In order to determine the optimal level of the parameters of the algorithms, the Taguchi method was used in the MINITAB 17 software. To use this method, the parameters are set at different levels depending on the type of algorithm used. According to the number of parameters of the algorithm and their different levels, the Taguchi L27 design (27 experiments) was employed for the MOICA and MOPSO algorithms. Tables 4 and 5 show the results of setting the parameters of the MOICA and MOPSO algorithms, respectively.

Table 4.

Setting the Parameters of the MOICA Algorithm.

Algorithm parameter	Parameter level
Algorithm parameter	Level 1	Level 2	Level 3	Optimal level
N-Pop	50	100	150	100
N-Emp	6	8	10	10
$α A$	.5	1	1.5	1.5
$β B$	1	2	3	3
$ξ$	.1	.15	.2	.15
$P_{A}$	0.4	0.6	0.8	0.8
$P_{R}$	0.2	0.3	0.4	0.3

Table 5.

Setting the Parameters of the MOPSO Algorithm.

Algorithm parameter	Parameter level
Algorithm parameter	Level 1	Level 2	Level 3	Optimal level
N-Pop	50	100	150	100
$C 1$	0.5	0.75	1	0.75
$C 2$	1	1.5	2	1.5
$W$	0.5	0.75	1	1
N-Rep	50	100	150	50
N-Grid	3	4	5	4

Comparison Indexes

The performance of multi-objective algorithms is much more complicated than the performance of single-objective algorithms, therefore, an evaluation index cannot be sufficient to investigate the results of the presented algorithms (Hajghani et al., 2023; Heidari et al., 2023; Kebriyaii et al., 2021). In general, a solution provided by multi-objective algorithms should have three characteristics:

- The distance between the set of dominant solutions obtained by the algorithm with the Pareto optimal set should be minimal.

- The distribution of solutions in the set of Pareto solutions should be uniform.

- The obtained solutions cover a large part of the values of each objective function.

In order to compare algorithms, various indexes are used based one: (Hajghani et al., 2023; Heidari et al., 2021; Jolai et al., 2013; Kebriyaii et al., 2021):

(1) Quality Metric (QM): The quality metric is such that all the Pareto solutions obtained by each of the algorithms are taken into consideration together, then, a non-dominated operation is performed for all the obtained solutions. Finally, the quality of each algorithm is equal to the share of new Pareto solutions specific to that algorithm. Higher quality means better the performance of the algorithm.

(2) Mean Ideal Distance (MID): The value of this index is equal to the distance of the Pareto points of the algorithm from the ideal point. In this study, since the objective functions of the proposed model are minimization, the ideal point is considered equal to the minimum value of each objective function. This index can be calculated by using Equation 23:

MID = \frac{\sum_{i = 1}^{n} \sqrt{{(\frac{f_{1 i} - f_{1}^{best}}{f_{1, total}^{\max} - f_{1, total}^{\min}})}^{2} + {(\frac{f_{2 i} - f_{2}^{best}}{f_{2, total}^{\max} - f_{2, total}^{\min}})}^{2}}}{n}

(23)

In the above relation, n is equal to the number of Pareto points and $f_{i, total}^{\max}$ and $f_{i, total}^{\min}$ are equal to the maximum and minimum value of the objective functions among all the non-dominated objective functions of the compared algorithms, respectively. In the above equation, the coordinates of the ideal point are equal to $(f_{2}^{best}, f_{1}^{best}) .$ The lower MID value means better the performance of the algorithm.

(3) Diversification Metric (DM): This index shows the extent of Pareto solutions of an algorithm and is calculated by Equation 24. The higher the value of the DM index, the better the performance of the algorithm.

DM = \sqrt{{(\frac{\max f_{1 i} - \min f_{1 i}}{f_{1, total}^{\max} - f_{1, total}^{\min}})}^{2} + {(\frac{\max f_{2 i} - \min f_{2 i}}{f_{2, total}^{\max} - f_{2, total}^{\min}})}^{2}}

(24)

(4) Spacing Metric (SM): This index indicates the uniformity of distributions of Pareto solutions in the solution space. The value of this index is computed by using Equation 25.

SM = \frac{\sum_{i = 1}^{n - 1} | \bar{d} - d_{i} |}{(n - 1) \bar{d}}

(25)

In the above equation, $d_{i}$ is equal to the Euclidean distance between two adjacent Pareto solutions in the solution space. Also, $\bar{d}$ is equal to the average distance of $d_{i}$ . The higher the value of the SM index, the better the performance of the algorithm.

Results

To examine the efficiency of the MOICA algorithm, random data with a specific probability distribution function is used. The data of the problem was generated by the MATLAB software presented in Table 6.

Table 6.

Problem Data.

Parameters	Value & distribution
n	10,20,30,40,50,60,70,80,90,100
$α$	.7
$γ$	.9
V	$~ U (1, 10)$
C	$~ U (1, 10)$
F	$~ U (200, 600)$
r	$~ U (1, 10)$
Chub	$~ U (50, 100)$
b	$~ U (30, 50)$
P	3
W	$~ U (0, 1)$

In this paper, to investigate the relative performance of the MOICA algorithm compared to the optimal front, the exact Goal Attainment method was used for the instance problems with the number of 10 and 20 nodes. It should be noted that larger problem sizes will have a longer solution time for this exact method. Tables show the comparison results. As it can be seen, the MOICA algorithm obtained close to optimal solutions. In these tables, the Pareto solutions are compared with each other with the Gap measure calculated by using Equation 26:

Gap = \frac{Ob j_{meta} - Ob j_{Optimal}}{Ob j_{Optimal}} \times 100

(26)

Tables 7 and 8 present the comparison of the Pareto solutions obtained by the MOICA algorithm with the solutions obtained by the Goal Attainment method for the problems with 10 nodes and 20 nodes, respectively.

Table 7.

Comparison of the Pareto Solutions Obtained by the MOICA Algorithm with the Goal Attainment Method for the Problem with 10 Nodes.

Pareto solution no.	Algorithm
	MOICA		Goal attainment		Gap (%)
	Obj 1	Obj 2	Obj 1	Obj 2	Obj 1	Obj 2
1	48,410.7	0.2701	48,410.7	0.2701	0	0
2	44,431.8	0.54813	44,361.01	0.54181	016	1.17
3	196,579.9	0.092832	195,421.9	0.092832	0.59	0

Table 8.

Comparison of the Pareto Solutions Obtained by the MOICA Algorithm with the Goal Attainment Method for the Problem with 20 Nodes.

Pareto solution no.	Algorithm
	MOICA		Goal attainment		Gap (%)
	Obj 1	Obj 2	Obj 1	Obj 2	Obj 1	Obj 2
1	233,821	1.5943	233,821	1.5943	0	0
2	269,327.5	1.528	269,327.5	1.5084	0	1.3
3	212,108.2	1.9433	211,984.3	1.9433	0.06	0
4	221,278.3	1.888	221,278.3	1.888	0	0
5	308,851.6	1.2912	306,743.5	1.2690	0.69	1.75

Figures 4 and 5 depicts one of the Pareto solutions obtained by the MOICA algorithm for the instance problems with 10 and 20 nodes, respectively.

Figure 4.

The Pareto solution obtained for the designed hub network problem with 10 nodes.

Figure 5.

The Pareto solution obtained for the designed hub network problem with 20 nodes.

In order to compare the efficiency of the multi-objective meta-heuristic algorithms, the instance problems with different sizes ranging from 10 to 100 nodes were solved by using both algorithms are executed and the results were compared by using the four aforementioned indexes. The comparison results of the MOPSO and MOICA algorithms are presented in Tables 9 and 10 as well as Figures 6 and 7.

Table 9.

Comparison Results of MOPSO and MOICA Algorithms Based on QM and SM Indexes.

Number of nodes	Number of hubs	Qauality metric (QM)		Spacing metric (SM)
Number of nodes	Number of hubs	MOPSO	MOICA	MOPSO	MOICA
10	2	0.2	1	0.510	0.592
20	4	0.138	0.642	0.744	0.445
30	7	0	1	0.513	0.615
40	9	0.1	0.9	0.727	0.461
50	11	0	1	0.605	0.354
60	15	0	1	0.578	0.410
70	18	0.1	0.9	0.659	0.419
80	24	0.2	0.8	0.835	0.260
90	26	0	1	0.650	0.565
100	28	0.5	0.9	0.504	0.406

Table 10.

Comparison Results of MOPSO and MOICA Algorithms Based on DM and MID Indexes.

Number of nodes	Number of hubs	Diversity metric (DM)		Mean ideal distance (MID)
Number of nodes	Number of hubs	MOPSO	MOICA	MOPSO	MOICA
10	2	0.822	0.881	0.573	0.470
20	4	1.042	0.920	0.705	0.592
30	7	0.750	1.200	0.720	0.455
40	9	1.080	1.207	0.760	0.330
50	11	0.952	1.358	0.890	0.562
60	15	0.780	1.341	0.609	0.273
70	18	1.010	1.101	0.630	0.391
80	24	0.788	1.135	0.605	0.406
90	26	0.804	0.832	0.505	0.501
100	28	1.039	0.804	0.587	0.593

Figure 6.

Comparison results of MOPSO and MOICA algorithms based on QM and SM indexes.

Figure 7.

Comparison results of MOPSO and MOICA algorithms based on DM and MID indexes.

As seen in the above tables and figures, the MOICA algorithm obtained better results than the MOPSO algorithm for all of the instance problems in terms of four indexes of QM, SM, DM, and MID.

Due to the fact that MOICA has performed better in terms of all indexes, the statistical analysis is performed to examine the significance/non-significance of this superiority. To evaluate the significant difference between the efficiency and performance of the algorithms proposed in this research, 10 problems have been considered. For this purpose, each algorithm was executed 10 times under the same conditions for each problem, and the average is reported as the solution for each problem in Tables 9 and 10. Then, the statistical analysis is conducted. In this study, the one-way ANOVA test for the equality of the mean of two populations is used. A decision should be made regarding the H₀ hypothesis based on the obtained value, that if α ≥ PVALUE, then the H₀ hypothesis (no significant difference between the averages of the algorithms) is rejected. However, if α ≤ PVALUE, there is no reason to reject H₀. In this test, α is assumed as .05.

As shown in Table 11, PVALUE is equal to .0392, which results in rejecting the H₀ hypothesis, hence, there is a significant difference between the performances of two algorithms in terms of the DM index.

Table 11.

The Output of the One-Way ANOVA Test for the DM Index

ANOVA table
Source	SS	df	MS	F	Prob > F
Columns	0.14655	1	0.14655	4.95	0.0392
Error	0.53322	18	0.02962
Total	0.67977	19

According to Figure 8, it can be said that the MOICA algorithm is significantly better.

Figure 8.

The output of the MATLAB software for the comparison of MOPSO and MOICA in terms of the DM index.

The ANOVA test was performed for the MID, QM, and SM indexes, and the amounts of 0.0007, 0, and 0.002 were obtained for PVALUE, respectively (presented in Tables 12 –14 and Figures 9 –11), which show there are also differences between the means of two populations in terms of these indexes as well.

Table 12.

The Output of the One-Way ANOVA Test for the MID Index.

ANOVA table
Source	SS	df	MS	F	Prob > F
Columns	0.20221	1	0.20221	16.64	0.0007
Error	0.21874	18	0.01215
Total	0.42095	19

Table 13.

The Output of the One-Way ANOVA Test for the QM Index.

ANOVA table
Source	SS	df	MS	F	Prob > F
Columns	3.12366	1	3.12366	165.21	1.65682e-10
Error	0.34033	18	0.01891
Total	3.46399	19

Table 14.

The Output of One-Way ANOVA Test for the SM Index.

ANOVA table
Source	SS	df	MS	F	Prob > F
Columns	0.16164	1	0.16164	13.02	0.002
Error	0.2234	18	0.01241
Total	0.38504	19

Figure 9.

The output of the MATLAB software for the comparison of MOPSO and MOICA in terms of the MID index.

Figure 10.

The output of the MATLAB software for the comparison of MOPSO and MOICA in terms of the QM index.

Figure 11.

The output of the MATLAB software for the comparison of the MOPSO and MOICA in terms of SM index.

Case Study

In this section, Digikala Group has been considered as a case study which is an online store in Iran and has more than 5.4 million visitors daily. More than 9.7 million different goods have been offered for sale by 308 thousand suppliers and it has the capacity to process 904 thousand goods per day. Digikala Group includes different departments and collections, some of which are operating inside the platform and some with different topics outside the Digikala platform.

In this case study, 38 cities in three provinces of Iran (Tehran, Hamadan, and Kermanshah) were considered, which have multiple orders. If this company seeks to build a hub to transfer orders to different customers, it will cost between 80,000 and 100,000 U.S. Dollars. The daily flows between hubs and customers usually are between 500 and 1,000. The transportation cost is based on the distance between the cities, which is between 1 and 3 U.S. Dollars for each unit. The hub capacity is between 20,000 and 40,000 units. The discount factor between two hubs is equal to 0.05 and between hub and non-hub nodes is equal to 0.01.

The proposed model was implemented on the case study and the Pareto points are provided in Table 15 and Figure 12. The results show that in order to improve the second objective function value, more hubs should be built, which lead to higher costs. On the other hand, in order to reduce cost, the second objective function should be ignored.

Table 15.

Pareto frontier related to the case study.

Row	Z1	Z2
1	4.06E+07	1.988
2	4.00E+07	2.157
3	3.46E+07	2.699
4	2.76E+07	2.777
5	2.36E+07	3.067

Figure 12.

Pareto frontier related to the case study.

The results for the Pareto point 2 show that hub number 15 will be built in Shahriar city located in Tehran province, the next hub will be built in Asadabad city located in Hamadan province and finally hub number 32 will be built in Kermanshah city, the capital of Kermanshah province. The allocation and flow of materials are presented in Table 16.

Table 16.

Allocation and Flow Between Origin and Destination.

Customer	Hub	Hub
Customer	Hub	15	17	32
1	15	9,723
2		9,582
3		9,980
4		10,484
5		7,100
6		10,522
7		10,686
8		7,216
9		9,738
10		7,538
11		11,353
12		7,013
13		10,036
14		10,900
15		150,033	76,422
16		10,274
17	17		9,578	140,462
18			6,775
19			7,603
20			8,253
21			8,048
22			10,771
23			7,514
24			6,621
25			11,259
26	32	10,254
27		11,234
28		10,827
29		10,920
30		11,608
31		10,450
32		10,844
33		9,829
34		11,153
35		10,740
36		10,756
37		10,898
38		10,949

Conclusion

In this research, the literature on HLPs was reviewed. Despite extensive studies on HLPs, there is a lack of sufficient research on problems applicable to the real world. On the other hand, there is an attempt to optimize different problems from different aspects, so that dealing with only one aspect may neutralize other essential aspects in decision-making.

It is obvious that quality criteria are effective in real-world HLPs. Another contribution of the research is the use of four qualitative criteria of service quality, traffic of the covered area, transportation infrastructure and expandability, along with the quantitative criteria that have been used, to calculate the weights of different nodes and using these weights in the objective function of the proposed bi-objective covering HLP.

Also, in most organizations, managerial decision-makers are interested in taking different scenarios into account to choose optimal conditions, so that one scenario cannot be considered better than the other. In other words, many problems have multiple goals that sometimes conflict with each other. Therefore, by considering different objective functions at the same time, various solutions can be presented so that decision-makers can choose the right solution with comparing the existing conditions and different aspects of the problem. Therefore, in this study, the bi-objective single-allocation coverage HLP with limited capacity was investigated in order to minimize costs and environmental risks.

The literature survey revealed that mainly quantitative criteria have been considered for the establishment of hubs in nodes, while in the real world, qualitative criteria are also effective in the location and manner of establishment of hubs. Therefore, another contribution of this research was the use of four aforementioned qualitative criteria along with the quantitative criteria that have been always addressed for calculating the node weights and using these weights in a separate objective function in a bi-objective maximal coverage hub location problem. In addition, due to the fact that in the investigated problem, production facilities (affiliated factories) were built in the hubs to reduce costs, evaluating the environmental risks of the nodes and turning this evaluation into a value to affect the nodes that became hubs were necessary, which was examined in the present study for the first time.

With the growth of population, social welfare, and economic development in recent years, the increase in transportation demand is inevitable. Also, in recent years, there has been growing attention to environmental issues because of the importance of reducing environmental pollution and environmental and social pressures. Transportation sector has destructive impact on the environment such as noise pollution and production of greenhouse gas emission. Hence, the optimization of energy consumption and reduction of environmental pollution in the transportation sector is necessary for continuing affordable services and reducing fuel consumption. A fundamental solution to reduce energy consumption in the transportation of goods is the correct design of transportation networks. In this research, the HLP was investigated to design an efficient hub network. The model was solved by using the Goal Programming method considering small-sized problems for validation. Also, the medium and large size problems were solved using the MOICA and MOPSO meta-heuristic algorithms. In addition, the performance of two algorithms were compared according to the QM, MID, DM, and SM indexes the results indicated that the MOICA algorithm outperformed the MOPSO algorithm for solving the problem in hand.

The following suggestions can be made for further studies: Instead of single allocation, multiple allocation can be considered for the proposed model. In addition, the route limitation may be taken into account. Also, deterministic problem data such as flow rate and cost can be considered non-deterministic. Moreover, the hub coverage problem can be modeled dynamically for different time periods. Furthermore, disruption should be taken into account in the HLPs.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Mohammad Khalilzadeh

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.

References

Abbasi-Parizi

Aminnayeri

Bashiri

(2018). Robust solution for a minimax regret hub location problem in a fuzzy-stochastic environment. Journal of Industrial & Management Optimization, 14(3), 1271.

Abdinnour-Helm

(2001). Using simulated annealing to solve the p-hub median problem. International Journal of Physical Distribution & Logistics Management, 31(3), 203–220.

Abdinnour-Helm

Venkataramanan

M. A.

(1998). Solution approaches to hub location problems. Annals of Operations Research, 78, 31–50.

Akgün

İ.

Tansel

B. Ç

. (2018). P-hub median problem for non-complete networks. Computers & Operations Research, 95, 56–72.

Alumur

S. A.

Campbell

J. F.

Contreras

Kara

B. Y.

Marianov

O’Kelly

M. E.

(2021). Perspectives on modeling hub location problems. European Journal of Operational Research, 291(1), 1–17. https://doi.org/10.1016/j.ejor.2020.09.039

Alumur

S. A.

Nickel

Saldanha-da-Gama

(2012). Hub location under uncertainty. Transportation Research Part B: Methodological, 46(4), 529–543.

Amin-Naseri

M. R.

Yazdekhasti

Salmasnia

(2018). Robust bi-objective optimization of uncapacitated single allocation p-hub median problem using a hybrid heuristic algorithm. Neural Computing and Applications, 29(9), 511–532.

Brimberg

Mišković

Todosijević

Uroševic

(2022). The uncapacitated r-allocation p-hub center problem. International Transactions in Operational Research, 29(2), 854–878.

Calık

Alumur

S. A.

Kara

B. Y.

Karasan

O. E.

(2009). A tabu-search based heuristic for the hub covering problem over incomplete hub networks. Computers & Operations Research, 36(12), 3088–3096.

10.

Chen

J. F.

(2007). A hybrid heuristic for the uncapacitated single allocation hub location problem. Omega, 35(2), 211–220.

11.

Chobar

A. P.

Adibi

M. A.

Kazemi

(2022). Multi-objective hub-spoke network design of perishable tourism products using combination machine learning and meta-heuristic algorithms. Environment, Development and Sustainability. https://doi.org/10.1007/s10668-022-02350-2

12.

Contreras

Cordeau

J. F.

Laporte

(2011). Stochastic uncapacitated hub location. European Journal of Operational Research, 212(3), 518–528.

13.

Cunha

C. B.

Silva

M. R.

(2007). A genetic algorithm for the problem of configuring a hub-and-spoke network for a LTL trucking company in Brazil. European Journal of Operational Research, 179(3), 747–758.

14.

de Camargo

R. S.

Miranda

Jr. Ferreira

R. P. M.

Luna

H. P

. (2009). Multiple allocation hub-and-spoke network design under hub congestion. Computers & Operations Research, 36(12), 3097–3106.

15.

Eydi

Nasiri

(2019). A p-hub median network design problem with preventive reliability approach for before and after hub failure. Sādhanā, 44(3), 57.

16.

Fattahi

Khalilzadeh

(2018). Risk evaluation using a novel hybrid method based on FMEA, extended MULTIMOORA, and AHP methods under fuzzy environment. Safety science, 102, 290–300.

17.

Gavriliouk

E. O.

(2009). Aggregation in hub location problems. Computers & Operations Research, 36(12), 3136–3142.

18.

Ghaffarinasab

Kara

B. Y.

Campbell

J. F.

(2023). The stratified p-hub center and p-hub maximal covering problems. Transportation Research Part B: Methodological, 157, 120–148. https://doi.org/10.1016/j.trb.2022.01.002

19.

Grine

F. Z.

Kamach

Khatab

Sefiani

(2021). An artificial immune system algorithm for solving the uncapacitated single allocation p-hub median problem. International Journal of Electrical & Computer Engineering, 11(3), 2293–2306.

20.

Hajghani

Forghani

M. A.

Heidari

Khalilzadeh

Kebriyaii

(2023). A two-echelon location routing problem considering sustainability and hybrid open and closed routes under uncertainty. Heliyon, 9(3), e14258. https://doi.org/10.1016/j.heliyon.2023.e14258

21.

Heidari

Imani

D. M.

Khalilzadeh

(2021). A hub location model in the sustainable supply chain considering customer segmentation. Journal of Engineering, Design and Technology, 19(6), 1387–1420. https://doi.org/10.1108/JEDT-07-2020-0279

22.

Heidari

Imani

D. M.

Khalilzadeh

Sarbazvatan

(2023). Green two-echelon closed and open location-routing problem: Application of NSGA-II and MOGWO metaheuristic approaches. Environment, Development and Sustainability, 25(9), 9163–9199. https://doi.org/10.1007/s10668-022-02429-w

23.

Hosseini

S. M.

Al Khaled

(2014). A survey on the imperialist competitive algorithm metaheuristic: Implementation in engineering domain and directions for future research. Applied Soft Computing, 24, 1078–1094. https://doi.org/10.1016/j.asoc.2014.08.024

24.

Ilić

Urošević

Brimberg

Mladenović

(2010). A general variable neighborhood search for solving the uncapacitated single allocation p-hub median problem. European Journal of Operational Research, 206(2), 289–300.

25.

Jolai

Asefi

Rabiee

Ramezani

(2013). Bi-objective simulated annealing approaches for no-wait two-stage flexible flow shop scheduling problem. Scientia Iranica, 20(3), 861–872.

26.

Kara

B. Y.

Tansel

B. C.

(2000). On the single-assignment p-hub center problem. European Journal of Operational Research, 125(3), 648–655.

27.

Kara

B. Y.

Tansel

B. C.

(2003). The single-assignment hub covering problem: Models and linearizations. Journal of the Operational Research Society, 54(1), 59–64.

28.

Kaushal

A. K.

Devi

(2023). Design of humanitarian logistics network using imperialist competitive algorithm. In: Pathak

Dubey

R. S.

(Eds.), Climate change and urban environment sustainability. Disaster resilience and green growth (pp. 227–264). Springer. https://doi.org/10.1007/978-981-19-7618-6_13

29.

Kebriyaii

Heidari

Khalilzadeh

Antucheviciene

Pavlovskis

(2021). Application of three metaheuristic algorithms to time-cost-quality trade-off project scheduling problem for construction projects considering time value of money. Symmetry, 13(12), 2402. https://doi.org/10.3390/sym13122402

30.

Kennedy

Eberhart

(1995). Particle swarm optimization [Conference session]. Proceedings of ICNN’95-International Conference on Neural Networks, Perth, Western Australia. IEEE.

31.

Klincewicz

J. G.

(1991). Heuristics for the p-hub location problem. European Journal of Operational Research, 53(1), 25–37.

32.

Klincewicz

J. G.

(1992). Avoiding local optima in the p-hub location problem using tabu search and grasp. Annals of Operations research, 40(1), 283–302.

33.

Fang

(2020). Robust hub location problem with flow-based set-up cost. IEEE Access, 8, 66178–66188.

34.

Maharjan

Hanaoka

(2020). A credibility-based multi-objective temporary logistics hub location-allocation model for relief supply and distribution under uncertainty. Socio-Economic Planning Sciences, 70, 100727.

35.

Maleki

Majlesinasab

Sinha

A. K.

(2023). An efficient model for the multiple allocation hub maximal covering problem. Optimization Methods and Software, 38(5), 1009–1030. https://doi.org/10.1080/10556788.2023.2196726

36.

Mumtaz

Eriskin

(2021). The minimal covering location and sizing problem in the presence of gradual cooperative coverage. European Journal of Operational Research, 295(3), 838–856. https://doi.org/10.1016/j.ejor.2021.03.015

37.

Nourzadeh

Ebrahimnejad

Khalili-Damghani

Hafezalkotob

(2020). Chance constrained programming and robust optimization approaches for uncertain hub location problem in a cooperative competitive environment. Scientia Iranica, 29(4), 2149–2165. https://doi.org/10.24200/SCI.2020.54072.3573

38.

O’Kelly

M. E.

(1987). A quadratic integer program for the location of interacting hub facilities. European Journal of Operational Research, 32(3), 393–404.

39.

Pamuk

F. S.

Sepil

(2001). A solution to the hub center problem via a single-relocation algorithm with tabu search. IEEE Transactions, 33(5), 399–411.

40.

Parvaresh

Husseini

S. M.

Golpayegany

S. H.

Karimi

(2014). Hub network design problem in the presence of disruptions. Journal of Intelligent Manufacturing, 25(4), 755–774.

41.

Weng

(2009). Path relinking approach for multiple allocation hub maximal covering problem. Computers & Mathematics with Applications, 57(11–12), 1890–1894.

42.

Rahmanniyay

Razmi

A. J.

(2020). An interactive multi-objective fuzzy linear programming model for hub location problems to minimise cost and delay time in a distribution network. International Journal of Logistics Systems and Management, 37(1), 79–105.

43.

Rahmati

Neghabi

(2021). Adjustable robust balanced hub location problem with uncertain transportation cost. Computational and Applied Mathematics, 40(1), 1–28.

44.

Rajabi

Avakh Darestani

(2020). Bi-objective optimization of a multi-product multi-period fuzzy possibilistic capacitated hub covering problem: NSGA-II and NRGA solutions. Journal of Optimization in Industrial Engineering, 13(2), 73–87.

45.

Rostami

Kämmerling

Naoum-Sawaya

Buchheim

Clausen

(2021). Stochastic single-allocation hub location. European Journal of Operational Research, 289(3), 1087–1106.

46.

Sasaki

Suzuki

Drezner

(1999). On the selection of hub airports for an airline hub-and-spoke system. Computers & Operations Research, 26(14), 1411–1422.

47.

Shang

Jia

Yang

Yuan

(2021). A credibility-based fuzzy programming model for the hierarchical multimodal hub location problem with time uncertainty in cargo delivery systems. International Journal of Machine Learning and Cybernetics, 12, 1413–1426.

48.

Shang

Yang

Jia

Gao

(2021a). Distributionally robust cluster-based hierarchical hub location problem for the integration of urban and rural public transport system. Computers & Industrial Engineering, 155, 107181.

49.

Shang

Yang

Jia

Gao

(2021b). The stochastic multi-modal hub location problem with direct link strategy and multiple capacity levels for cargo delivery systems. Transportmetrica A: Transport Science, 17(4), 380–410.

50.

Sim

Lowe

T. J.

Thomas

B. W.

(2009). The stochastic p-hub center problem with service-level constraints. Computers & Operations Research, 36(12), 3166–3177.

51.

Skorin-Kapov

O’Kelly

(1996). Tight linear programming relaxations of uncapacitated p-hub median problems. European Journal of Operational Research, 94(3), 582–593.

52.

Snyder

H. N.

Sickmund

(2006). Juvenile Offenders and Victims: National Report. Office of Juvenile Justice and Delinquency Prevention.

53.

Soleimani

Khalilzadeh

Bahari

Heidary

(2022). NSGA-II algorithm for hub location-allocation problem considering hub disruption and backup hub allocation. World Journal of Engineering, 19(6), 794–807. https://doi.org/10.1108/WJE-12-2020-0658

54.

Talbi

E. G.

Todosijević

(2017). The robust uncapacitated multiple allocation p-hub median problem. Computers & Industrial Engineering, 110, 322–332.

55.

Tang

Dai

Salido

M. A.

Giret

(2016). Energy-efficient dynamic scheduling for a flexible flow shop using an improved particle swarm optimization. Computers in Industry, 81, 82–95.

56.

Tikani

Ramezanian

Setak

Van Woensel

(2021). Hybrid evolutionary algorithms and Lagrangian relaxation for multi-period star hub median problem considering financial and service quality issues. Engineering Applications of Artificial Intelligence, 97, 104056.

57.

Tofighian

Arshadi khamseh

(2020). A Bi objective uncapacitated multiple allocation p-hub median problem in public administration considering economies of scales. Research in Transportation Economics, 90, 100896.

58.

Topcuoglu

Corut

Ermis

Yilmaz

(2005). Solving the uncapacitated hub location problem using genetic algorithms. Computers & Operations Research, 32(4), 967–984.

59.

Vahdani

Mohammadi

(2015). A bi-objective interval-stochastic robust optimization model for designing closed-loop supply chain network with multi-priority queuing system. International Journal of Production Economics, 170, 67–87.

60.

Vural

Aygün

(2019). Capacited p-hub location problem allowing direct flow between spokes in intermodal transportation network. Sādhanā, 44(9), 1–11.

61.

Wang

Zhou

(2024). A lattice Boltzmann method-like algorithm for the maximal covering location problem on the complex network: Application to location of railway emergency-rescue spot. Mathematics, 12(2), 218. https://doi.org/10.3390/math12020218

62.

Wang

Qin

(2021). Uncertain programming models for sustainable hub covering location problem over incomplete network. IEEE Access, 9, 16092–16102.

63.

Weng

Yang

(2006). Two artificial intelligence heuristics in solving multiple allocation hub maximal covering problem [Conference session]. Intelligent Computing. ICIC 2006, Kunming, Yunnan, China. Lecture Notes in Computer Science.

64.

Yaman

Elloumi

(2012). Star p-hub center problem and star p-hub median problem with bounded path lengths. Computers & Operations Research, 39(11), 2725–2732.

65.

Yang

Liu

Yang

(2013). An improved hybrid particle swarm optimization algorithm for fuzzy p-hub center problem. Computers & Industrial Engineering, 64(1), 133–142.

66.

Yang

Liu

Yang

(2014). Optimizing fuzzy p-hub center problem with generalized value-at-risk criterion. Applied Mathematical Modelling, 38(15–16), 3987–4005.

67.

Yang

T. H.

(2009). Stochastic air freight hub location and flight routes planning. Applied Mathematical Modelling, 33(12), 4424–4430.

68.

Yin

Chen

Song

Liu

(2019). A new distributionally robust p-hub median problem with uncertain carbon emissions and its tractable approximation method. Applied Mathematical Modelling, 74, 668–693.

69.

Zahedi

Ghahremani Nahr

(2020). Designing a hub covering location problem under uncertainty conditions. Management Science Letters, 9(3), 477–500.

A Bi-Objective Mathematical Programming Model for a Maximal Covering Hub Location Problem Under Uncertainty

Abstract

Keywords

Introduction

Literature Review

Reviewing the Hub Location Problems (HLPs)

Reviewing the Solution Approaches

Problem Definition

Proposed Mathematical Programming Model

Mathematical Programming Model

Solution Approach

Goal Attainment or Chebyshev Goal Programming Method

The Multi-Objective Imperialist Competitive Algorithm (MOICA) Meta-Heuristic Algorithm

The Multi-Objective Particle Swarm Optimization (MOPSO) Algorithm

Solution Representation Method

Setting Parameters by the Taguchi Method

Comparison Indexes

Results

Case Study

Conclusion

Footnotes

Declaration of Conflicting Interests

Funding

ORCID iD

Data Availability Statement

References