INTERNATIONAL

TRANSACTIONS
IN OPERATIONAL
Intl. Trans. in Op. Res. 00 (2018) 1–22 RESEARCH
DOI: 10.1111/itor.12566

Bi-criteria formulation for green scheduling with unrelated

parallel machines with sequence-dependent setup times
Luciano P. Cotaa,b , Vitor N. Coelhoc , Frederico G. Guimarãesd and
Marcone J. F. Souzae
a
Graduate Program in Electrical Engineering, Universidade Federal de Minas Gerais,
Belo Horizonte, MG 31270-901, Brazil
b
Vale Institute of Technology, ITV Mining, Ouro Preto, MG 35400-000, Brazil
c
Institute of Computer Science, Universidade Federal Fluminense, Niterói, RJ 24210-346, Brazil
d
Department of Electrical Engineering, Universidade Federal de Minas Gerais, Belo Horizonte, MG 31270-010, Brazil
e
Department of Computer Science, Universidade Federal de Ouro Preto, Ouro Preto, MG 35400-000, Brazil
E-mail: [email protected] [Cota]; [email protected] [Coelho]; [email protected] [Guimarães];
[email protected] [Souza]
Received 10 January 2017; received in revised form 15 May 2018; accepted 17 May 2018

Abstract
Given the important role of machine scheduling in manufacturing industry, we discuss power consumption
in sequencing jobs in a scheduling problem, assuming variable speed operation in machines. The problem
involves defining the allocation of jobs to machines, the order of processing jobs and the speed of processing
each job in each machine. This problem can be viewed as a type of green scheduling problem, dealing with
sustainable use of energy consumption and environmental effects. We propose a mixed integer linear program-
ming (MILP) model for the unrelated parallel machine-scheduling problem with sequence-dependent setup
times, with independent and non-preemptible jobs, minimizing the makespan and the total consumption
of electricity. Furthermore, we employ a novel math-heuristic algorithm, named multi-objective smart pool
search matheuristic (or simply smart pool), for finding solutions near the Pareto front, in a restricted compu-
tational budget. As a case study, a new set of instances is created for the problem. Those instances are solved
using the classical -constrained method and the smart pool method. The obtained sets of non-dominated
solutions indicate the conflict between both objectives, highlighting the relevance of the suggested approach
to industry. From the obtained results, it was verified that the smart pool achieved good convergence towards
the true Pareto front, as indicated by the hyper-volume metric, presenting lower average time for finding
solutions on the Pareto front. In small to medium size instances, the smart pool search method can achieve
very good approximations of the Pareto front with less computational effort than traditional methods.

Keywords: green scheduling; multi-objective optimization; parallel machines; mixed integer linear programming


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
Published by John Wiley & Sons Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main St, Malden, MA02148,
USA.
2 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
1. Introduction

In the manufacturing sector, the predominant energy sources are natural gas and electricity. Ac-
cording to the U.S. Energy Information Administration, the industrial sector consumes about 54%
of the world’s total delivery energy. Energy is used in the industrial sector for a wide range of pur-
poses, such as process and assembly, steam and cogeneration, process heating and cooling, lighting,
heating and air conditioning for buildings. The energy consumption in industrial activities is also
related to the emission of greenhouse gases in the atmosphere.
Given the growing awareness and demand for sustainability and rational use of natural resources,
the research for novel techniques for an efficient use of materials and energy resources has intensified
in the recent years (Mansouri et al., 2016). These research directions involve the study of materials
reuse, recycling, lower energy consumption, lower generation of pollutants, conscious use of water
resources, decentralization of the energy system (Coelho et al., 2017b) and investments on micro-
grids powered by renewable energy sources (Coelho et al., 2016b), among others. Sustainable
development relies on the efficient use of energy as well as development of energy management
systems.
In Brazil, 44% of electricity consumption is in the industrial sector (Sauer et al., 2015). Moreover,
three-phase induction electrical motors account for 68% of the industrial consumption (Sauer
et al., 2015), motors are responsible for 35% of the electrical energy consumption in Brazil. This
figure is higher in many developed countries, with electricity consumption due to motors ranging
from 50% in United Kingdom to 80% in Canada. The energy generation in Brazil is composed
mainly of hydroelectric dams and thermoelectric plants to supply additional energy when needed.
Despite the increasing use of distributed generation in the last years, the participation of renewable
energy sources is still timid, below 5% of national production of energy. Thus, reducing total energy
consumption (TEC) in industry is paramount in order to decrease greenhouse gase emissions and
environmental impact and also indirectly reduce production costs in industry.
Machine scheduling plays an important role in manufacturing industry, allowing production
managers and production engineers to maximize productivity by allocating the jobs to the resources
available in an optimal way. In this paper, we discuss electric energy consumption in sequencing
jobs in a machine scheduling problem. We propose a multi-objective formulation in which one of
the objectives is to minimize the TEC in the scheduling while processing all the required jobs. In
particular, the variant called unrelated parallel machine scheduling problem with setup times is
addressed. We select this problem due to its practical application in industry, since it is a reality in
many industrial processes of different companies such as textile, chemical, semiconductor, inks and
papers (Franca et al., 1996; Randhawa and Kuo, 1997; Radhakrishnan and Ventura, 2000; Kurz
and Askin, 2001; Rabadi et al., 2006).
Formally, the unrelated parallel machine scheduling problem with setup times involves a set
N = {1, . . . , n} of independent and non-preemptible jobs and a set M = {1, . . . , m} of machines.
The job processing times and setup times are dependent on the sequence. In this work, we focus on
allocating all the jobs to the machines seeking to minimize the makespan (the time to finish the last
job in the schedule) and the total consumption of electricity by machines. The scheduling problem
considering only the minimization of makespan is usually represented by RM |Si jk |Cmax , according to
the three-field standard notation introduced by Graham et al. (1979), where RM indicates unrelated
machines, Si jk indicates that the setup times are sequence-dependent and Cmax defines the makespan.


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 3
We introduce in this paper the bi-objective variant herein defined as RM |Si jk |(Cmax , TEC), where
T EC represents the total consumption of electricity spent by machines as an additional objective.
The model will be further detailed in Section 3.
The proposed formulation RM |Si jk |(Cmax , TEC) has the following characteristics: (a) each job
should be allocated to only one machine;1 (b) each job has a processing time to complete in a
machine; (c) there is a setup time to calibrate each machine to process a job. The setup time depends
on the order of allocation of jobs in the machine; (d) each machine has a given power consumption
according to speed of operation to process the job in the machine and (e) there is a discrete set
of processing speeds to process a job in a machine. In this problem, the objective is to allocate
all jobs in the machines, to define their sequences on machines and to find their processing speed,
minimizing the makespan and the total consumption of electricity.
The objectives makespan and total consumption of electricity have great importance in this prob-
lem. The minimization of the makespan usually implies in good use of the machines available in the
production (Pinedo, 2008). Additionally, the minimization of the total consumption of electricity
implies in reduction of costs for the industries and the conscious use of environmental resources.
The conflicting nature of the objectives Cmax and TEC is due to the trade-off between maximiz-
ing production at higher speeds of operation, which would lead to greater power consumption
in scheduling. The majority of electrical motors used in industry are induction machines, given
their high efficiency, the low maintenance costs and the variable-speed control allowed by modern
frequency inverters (variable frequency drive method). In variable speed applications, changing
frequency is a common method to control the speed of an induction motor, particularly with the
available power electronic devices, a variable supply frequency to the stator can be used to con-
trol speeds. However, we have to make sure that the voltage–frequency ratio remains constant to
maintain constant flux in the machine. Therefore, the use of a machine in a greater speed leads to a
corresponding increase in power input (Sen, 2013; Wildi, 2013). These characteristics are assumed
in the formulation RM |Si jk |(Cmax , TEC).
Research involving sustainability in scheduling problems have been called green scheduling (Yu,
2010; Bampis et al., 2015; Mansouri and Aktas, 2016; Mansouri et al., 2016). In recent papers, the
makespan has been minimized with other objectives related to consumption of energy for other
machine scheduling problems. Mansouri et al. (2016) and Mansouri and Aktas (2016) minimize the
makespan and the total consumption of energy in different problems. Also, Wang et al. (2016) and
Cheng et al. (2017) minimize the makespan and the total energy costs in a different formulation.
The problem tackled here has great importance, because it is present in production processes
of many industries (Rabadi et al., 2006; Pereira Lopes and de Carvalho, 2007) and belongs to
the N P -hard class of problems, since it is a generalization of the parallel machine scheduling
problem with identical machines and without setup times (Karp, 1972; Garey and Johnson, 1979).
Machine scheduling problems with setup times have been attracting attention of several researchers,
because of its importance to the industry and the challenges imposed for its resolution (Zhu and
Wilhelm, 2006). These problems can be found, for example, in furniture manufacturing (Agnetis
et al., 2001) and paint factories (Mansouri, 2005). Several works in the literature dealt with the
unrelated parallel machine scheduling problem with setup times, but few approaches had analysed

1
Except the fictitious job 0, which is usually allocated at the beginning of each machine in the implementations of
mathematical models.


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
4 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
energy consumption. In this sense, this present work contributes with a more sustainable view
regarding classical machine schedules. For this purpose, a mixed integer linear programming (MILP)
mathematical model is designed, implemented and analysed. In particular, the proposed model is
inspired from the work of Mansouri et al. (2016), in which the authors propose a mathematical
model for green scheduling in a flow shop problem.
In order to tackle the proposed MILP model and generate good quality sets of non-dominated so-
lutions, in a restricted computational time, we use the multi-objective smart pool search matheuristic
(or simply smart pool for short), recently proposed by Coelho et al. (2016a). The main core of this
method is to transform a multi-objective problem into multiple single-criterion weighted-sum prob-
lems. A new set of instances is designed in order to validate the proposal. Obtained solutions from
the classical -constrained method and the smart pool are presented and compared. The trade-off
between both objective functions is verified, showing the ability of the proposed framework to
provide efficient schedules with low energy consumption.
The rest of this paper is organized as follows. Section 2 presents a literature review. The proposed
mathematical model is introduced in Section 3. Smart pool method is detailed in Section 4. In
Section 5, the results on computational experiments are reported and discussed. Finally, in Section 6,
the conclusions and future work are presented.

2. Literature review

In Glass et al. (1994), the authors proposed a genetic algorithm, a simulated annealing (SA), and a
tabu search (TS) for a problem without setup time. A TS is proposed by Srivastava (1997) for the
same problem. In Weng et al. (2001), seven constructive heuristics are proposed to solve a problem
with machine-independent sequence-dependent setup times with the objective of minimizing the
total weighted completion time. In Kim et al. (2002, 2003), the authors proposed a SA to minimize
the total tardiness in a problem with machine-independent sequence-dependent setup times. In
Logendran et al. (2007), six different TS are proposed to minimize the total tardiness in a problem
with machine-independent sequence-dependent setup times.
In the following, we review some works that approached the problem of minimizing only the
makespan (defined as RM |Si jk |Cmax ). In Al-Salem (2004), the authors developed a partitioning
heuristic that combined a constructive heuristic, a local search strategy and a heuristic based on
the asymmetric traveling salesman problem. In the work of Rabadi et al. (2006), a metaheuristic
for randomized priority search (Meta-RaPS) is compared with the exact method applied to solve
the proposed mathematical model. An Ant Colony Optimization (ACO) algorithm is proposed by
Arnaout et al. (2010), for solving an special case of the problem where the distance between jobs
and machines is considerably large. In Ying et al. (2012), a restricted simulated annealing (RSA)
method is proposed, mainly based on an efficient elimination of inefficient jobs. Two different
genetic algorithm are proposed by Vallada and Ruiz (2011), as well as a new mathematical model.
A bee colony algorithm is developed by Lin and Ying (2014). In Avalos-Rosales et al. (2013), the
authors propose a new mathematical model built upon some ideas in the model of Vallada and
Ruiz (2011). This new model has better performance than the models of Rabadi et al. (2006) and
Vallada and Ruiz (2011). In the work of Arnaout et al. (2014), an improvement in the ACO of
Arnaout et al. (2010) is done. In the works of Haddad et al. (2014, 2015) and Cota et al. (2014a,

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 5
2014b), the authors proposed different algorithms that combine the iterated local search (Lourenço
et al., 2003) and variable neighbourhood descent (Hansen et al., 2008) metaheuristics, as well
as other techniques. The so-called AIRP algorithm, designed by Cota et al. (2014a), is the one
that presented better performance. In Avalos-Rosales et al. (2013, 2015), the authors proposed
improvements in the mathematical model presented in Vallada and Ruiz (2011). In Santos et al.
(2016), the authors presented an analysis of several stochastic local search methods, with previous
calibration of parameters. In Tran et al. (2016), the authors proposed two hybrid models based
on logic-based Benders decomposition and branch and check. An adaptive large neighbourhood
search with learning automata is proposed by Cota et al. (2017).
Sustainable use of electricity in scheduling problems is still under-explored in the literature. In
Zhang et al. (2014), a time-indexed mathematical model for solving flow shop scheduling problem
is proposed for dealing with two objectives: minimizing the electricity costs and carbon emissions.
This paper analysed the variation in the electricity costs during consumption time. In the work
of Ding et al. (2016), the authors dealt with another flow shop scheduling problem for minimiz-
ing makespan and total carbon emissions, using a multi-objective iterated greedy algorithm. In
Mansouri et al. (2016), a two-machine flow shop problem, with setup times, is tackled aiming
at minimizing makespan and TEC. For its resolution, a mathematical model and a constructive
heuristic were proposed for analysing trade-off between objectives.
In Wang et al. (2016), the authors treat the single machine batch scheduling problem with non-
identical job sizes, time-of-use (TOU) electricity prices and different energy consumption rates
of the machine. The objectives are to minimize the makespan and to minimize the total energy
costs. The authors proposed a mathematical model for the problem and implemented an exact
epsilon-constraint method. In Liu et al. (2016), the authors proposed a mathematical model for the
bi-objective optimization problem that minimizes the total non-processing electricity consumption
and total weighted tardiness in a job shop. For its resolution, they developed a multi-objective genetic
algorithm based on NSGA-II (Deb et al., 2002). A two-machine sequence-dependent permutation
flow shop problem with the objectives to minimize the makespan and the energy consumption is
treated in Mansouri and Aktas (2016). The authors proposed constructive heuristics and multi-
objective genetic algorithms for the resolution. A single-machine batch scheduling problem with
TOU electricity prices is treated in Cheng et al. (2017). The objectives of the problem are to minimize
the makespan and the total electricity cost. The authors proposed a mathematical model and an
-constraint method.
In this paper, we propose a multi-objective formulation for green scheduling, considering the
unrelated parallel machine scheduling problem with setup times with the objectives of minimizing
the makespan and the TEC. The objective TEC is very important because there is great concern
about the conscious use of energy in the present times, and these resources imply in high production
costs for industries. This work intends to encourage the use of green scheduling in the problem and
to provide a mathematical model that can be used in practical cases in the industry.

3. Mathematical model

In contrast to the classical approaches for the unrelated parallel machine scheduling problem with
setup times, as shown within the works of Rabadi et al. (2006), Vallada and Ruiz (2011) and

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
6 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
Avalos-Rosales et al. (2013, 2015), which only minimize makespan, new characteristics are added to
the proposed formulation. These features are necessary for the consideration of energy consumption,
such as the variable processing speed and the machine power at different speeds. The inclusion of
these characteristics is inspired by the work of Mansouri et al. (2016) for a flow shop scheduling
problem. The new mathematical model proposed in this work is also based on the one introduced
by Avalos-Rosales et al. (2015). It is worth mentioning that independent and non-preemptible jobs
are considered.
The machine scheduling problem addressed in this study (RM |Si jk |(Cmax , TEC)) has the following
detailed characteristics:

1. There are a set N of jobs, a set M of machines and a set L of different modes of operation, each
one related to a corresponding speed of operation and power consumption.
2. The machines i ∈ M are independent parallel, so the processing time of a job is different in each
of the machines.
3. All jobs are available to be processed starting from time 0.
4. Each job j ∈ N should be allocated to a single machine i ∈ M.
5. The time to process the job j ∈ N on a given machine i ∈ M is the processing time pi j .
6. There is a setup time Si jk to process the job k ∈ N after the job j ∈ N on the machine i ∈ M, in
this order.
7. Each machine i ∈ M has a power consumption πi at normal speed of operation.
8. Each mode of operation l ∈ L has a multiplication factor of speed vl at normal operation.
Thereby, the real processing time of the job k ∈ N on the machine i ∈ M and on mode operation
p
l is vik .
l
9. Each mode of operation l ∈ L has a multiplication factor of power λl at normal operation. This
factor is used to calculate the energy consumption on each mode of operation.

The goal of the problem involves the allocation of all the jobs (N) on the machines (M) using the
modes of operation (L), seeking to minimize two objectives. The first objective is to minimize the
makespan, which is the time to finish the last job in the schedule. The second objective is to minimize
the consumption of electricity, which is calculated using the power consumption, the multiplication
factor of power, the multiplication factor of speed and the processing time.
The notation of the new mathematical model is given as follows:
r M = {1, . . . , m}: set of machines, being m its cardinality;
r N = {1, . . . , n}: set of jobs with n representing the number of jobs;
r L = {1, . . . , q}: set of q different modes of operation, each mode is related to a corresponding
speed of operation and power consumption (later called just modes of operation);
r N0 = N ∪ {0}: set of jobs with inclusion of the fictitious job 0;
r pi j : processing time of job j in machine i [minutes];
r Si jk : setup time necessary for allocating job k in machine i after job j [minutes];
r B: big constant;
r πi : power consumption of machine i at normal speed of operation [kW];
r vl : multiplying factor of speed at normal operation, with l ∈ L;
r λl : multiplying factor of power at normal speed, with l ∈ L.

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 7
We assume that the factors vl and λl are the same on all machines. The values of the factors are
inspired in the study proposed in Ahilan et al. (2013), which is also used by Mansouri and Aktas
(2016) and Mansouri et al. (2016).
As an example, if the job j has processing time of 100 units of time in a certain machine and the
factor vl = 0.8, then the time to processing j becomes 125 units of time. If the factor is vl = 1.2,
then the time to processing j becomes 83.3. Similarly, machine speed and power influence the
energy consumption, thus in our model, we assume that the higher the speed, the greater the energy
consumption is. At normal speed of operation, vl = 1.0 and λl = 1.0, and then the processing time
of job j in machine k is given by p jk , while the power consumption is given by πk . The factor vl is a
non-decreasing function of λl . The relation between the factors vl and λl is given below:
⎧
⎪
⎪ vl = 1 and λl = 1, normal speed of machine operation,
⎪
⎪
⎨ 0 < vl < 1 and 0 < λl < 1, speed slower than normal, then the machine
vl and λl = consumes less power,
⎪
⎪
⎪
⎪ vl > 1 and λl > 1, speed greater than normal, then the machine
⎩ consumes more power.
Decision variables used in the mathematical model are

1, if job k, with mode of operation l, is allocated immediately after job j in machine i
xi jkl =
0, otherwise.
Auxiliary variables used in the model are as follows:
r C j : completion time of job j;
r Oi : completion time of machine i;
r Cmax : maximum processing time of all machines (makespan);
r TEC: total energy consumption (kWh).

This model uses a fictitious job 0 allocated at the beginning of each machine. This job has
processing times pi0 = 0 ∀i ∈ M and setup time Si0k = 0 ∀i ∈ M, ∀k ∈ N. The mathematical model
can be seen from Equations (1) to (12):

min Cmax (1)

min TEC (2)

Subject to:


m 
n 
q
xi jkl = 1 ∀k ∈ N (3)
i=1 j=0 l=1
j=k


m 
n 
q
xi jkl ≤ 1 ∀ j ∈ N (4)
i=1 k=1 l=1
j=k


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
8 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22


n 
q
xi0kl ≤ 1 ∀ i ∈ M (5)
k=1 l=1


n 
q

n 
q
xi jkl − xih jl = 0 ∀ j ∈ N, ∀i ∈ M (6)
k=0 l=1 h=0 l=1
k= j h= j

 p
Ck − C j + B 1 − xi jkl ≥ Si jk + ik ∀ j ∈ N0 , ∀k ∈ N,
vl
j = k, ∀l ∈ L,
∀i ∈ M (7)

C0 = 0 (8)
q


m 
n 
pik
Si jk + xi jkl = Oi ∀i ∈ M (9)
vl
j=0 k=1 l=1
k= j

Cmax ≥ Oi ∀i ∈ M (10)
q


m 
n 
n 
πi pik
TEC ≥ λl × × xi jkl (11)
60 vl
i=1 j=0 k=1 l=1
j=k

xi jkl ∈ {0, 1} ∀ j ∈ N0 , ∀k ∈ N,
j = k, ∀i ∈ M,
∀l ∈ L. (12)

The objective of the model is to minimize makespan (1) and TEC (2). Constraints (3) ensure
that each job will be allocated to only one machine, as well as having a predecessor, and working a
unique operation mode. Constraints (4) define that each job will have, at most, one successor job.
Analogously, in constraints (5) it is guaranteed that the dummy job has at most one successor job.
Constraints (6) ensure right order for allocating job, if a job j is predecessor to a job k, there must be
another job that comes before job j. Constraints (7) are responsible for calculating the accumulated
time of each job, if xi jkl = 1, then, accumulated time of k (Ck ) is equal to the one of j (C j ) plus
the sum of the time for preparing the machine (Si jk ) and the processing time of job k at mode of
p
operation l ( vik ). If xi jkl = 0, constant B will ensure the constraints are satisfied. In constraints (8),
l
the accumulated time of the fictitious job 0 is ensured. Those described in constraints (9) perform
calculations of the accumulated costs for each machine, this cost is given by the sum of the setup time
and the processing time of all jobs allocated to a given machine. Constraints (10) define the value of
Cmax . The calculation of the TEC is performed by constraint (11). To perform this calculation, one
needs to consider the processing times (pik /vl ), the power input of each machine at normal speed

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 9
Table 1
Setup times of machines M1 and M2

M1 1 2 3 4 5 6 M2 1 2 3 4 5 6
1 0 1 8 1 3 9 1 0 5 1 6 1 7
2 4 0 7 3 7 8 2 6 0 7 7 6 2
3 7 3 0 2 3 5 3 7 6 0 9 6 9
4 3 8 3 0 5 2 4 3 7 3 0 1 7
5 8 3 7 9 0 5 5 5 8 5 6 0 9
6 8 8 1 2 2 0 6 7 4 1 7 9 0

Fig. 1. Optimal solution for RM |Si jk |Cmax . [Colour figure can be viewed at wileyonlinelibrary.com]

of operation (given by πi ), and the multiplying factors (λl ) and (vl ). The machine power is divided
by 60 because the power is given in kilowatt, while the processing time is given in minutes and the
TEC is calculated in kilowatt hour. This constraint is equivalent to the integral calculation of the
energy consumption. Finally, constraints (11) define the binary variables.
The proposed mathematical model has n2 mq binary variables and n + m + 2 continuous variables.
The model also has 2n + 3m + nm + 2n2 mq + 2 constraints. Therefore, adding the consideration of
variable speed leads to a linear increase in the number of binary variables and constraints.
In order to illustrate this problem, we choose randomly an instance with six jobs, two machines and
one operation mode (normal speed of machine operation with vl = 1 and λl = 1). The processing
times and the power consumption at normal speed of operation of the machines M1 and M2 are
given below. Setup times of those machines are shown in Table 1:

M1 : p1 j = {1, 87, 28, 32, 38, 9}, π1 = 70,

M2 : p2 j = {4, 21, 68, 17, 43, 48}, π2 = 179.

This instance was solved with the IBM ILOG CPLEX, version 12.5. Figure 1 shows the optimal
solution for the problem RM |Si jk |Cmax by considering just the minimization of makespan. Jobs 1,
4, 6 and 3, in this order, are allocated to machine 1 and jobs 2 and 5, in this order, are allocated to
machine 2. Figure 2 shows the optimal solution for the problem RM |Si jk |TEC by minimizing just
the TEC. In this case, jobs 6, 4, 1, 3 and 5, in this order, are allocated to machine 1 and only jobs 2,
is allocated to machine 2.

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
10 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22

Fig. 2. Optimal solution for RM |Si jk |TEC. [Colour figure can be viewed at wileyonlinelibrary.com]

It can be observed that there is a great variability in the allocation and order of the jobs and the
values of the objectives. In Fig. 1, the makespan is 74 units of time and the TEC is 272.60 units
of energy. In contrast, in Fig. 2, the makespan is 124 units of time and the TEC is 188.65 units
of energy. This example shows the conflicting nature of the two objectives in the scheduling. An
interesting trade-off could be found in which the jobs are processed in different speeds in different
machines.

4. Multi-objective smart pool search matheuristic

The multi-objective smart pool search matheuristic method (or smart pool) (Coelho et al., 2016a)
has the goal of finding non-dominated solutions near the Pareto front. This matheuristic method
consists in solving multi-objective mathematical models by optimizations done by mathematical
solvers, within a predefined time limit for each resolution called by the procedure. In order to
achieve its goal, different mixed integer linear programming (MILP) problems are generated with
different weights for aggregating both objective functions involved in the proposed formulation.
Algorithm 1 presents the smart pool search pseudo-code.

Algorithm 1: Multi-objective smart pool search matheuristic


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 11
In this paper, the weights were generated using the method proposed by Scheffé (1958). This
proposition defines the {r, wmax }-simplex lattice, used in the experiments with mixtures.
 In the context
r+wmax −1
of optimization problems, r is the number of objectives. In this method, wmax
points in r-
dimensional space are generated, with wmax + 1 points equally spaced in the boundary of the simplex
and satisfying the condition: ||w||1 = w1 + w2 + · · · + wr = 1. This method for weight generation is
very popular in multi-objective evolutionary algorithms such as NSGA-III and MOEA/D (Zhou
et al., 2011; Deb and Jain, 2014). It is more efficient than the one used in the original version of the
smart pool (Coelho et al., 2016a). Other approaches for weight vector generation could have been
used too, see for instance Meneghini and Guimarães (2017).
In Algorithm 1, line 4 generates the single criteria (weighted sum aggregation) mathematical model
with desired weights w1 and w2 for the objectives makespan and TEC, respectively. Subsequently,
the line 5 calls the black-box solver for optimizing the model. During the search and optimization
of each MILP problem, usually, different feasible solutions can be found. In this sense, the pool of
feasible solutions obtained in each run of the optimizer are returned and stored in the set poolSol.
Parameter timeLim sets the maximum time that can be spent by the solver within the search of each
MILP problem with a unique set of weights.
Finally, line 8 checks which of these solutions are non-dominated, filtering poolSol with a simple
Pareto dominance, considering the procedure addSolution, detailed by Lust and Teghem (2010)
and depicted in Algorithm 2. The procedure keeps updating the set of non-dominated solutions
Xe , initially empty (line 2 of Algorithm 1), filtering the solutions extracted from the BB tree (or
other method used by the mathematical solver) and creating an approximation of the Pareto
front.

Algorithm 2: addSolution


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
12 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
Table 2
Characteristics of the problem instances

Parameters Levels Based on

Number of jobs (n): 6, 8, 10, 12, 15 Vallada and Ruiz (2011)
Number of machines (m): 2, 3, 4, 5 Vallada and Ruiz (2011)
Modes of operation (q): 3 or 5 Mansouri et al. (2016) and
Ahilan et al. (2013)
Processing time (pi j ): U [1, 99] Vallada and Ruiz (2011)
Sequence dependent setup time (Si jk ): U [1, 9], U [1, 49], U [1, 99], U [1, 124] Vallada and Ruiz (2011)
Machine power (πi ): U [40, 200] –
Multiplying factor of speed (vl ): 1.2, 1.1, 1, 0.9, 0.8 Mansouri et al. (2016) and
Ahilan et al. (2013)
Multiplying factor of power (λl ): 1.5, 1.25, 1, 0.8, 0.6 Mansouri et al. (2016) and
Ahilan et al. (2013)

5. Computational results

Computational experiments were done considering a computer with Core i7, 1.9 GHz, 6 GB RAM
and operational system Ubuntu 16.04. The mathematical model was implemented in the OptFrame
2.3 (Coelho et al., 2011) and solved with the IBM ILOG CPLEX, version 12.5.

5.1. Instances generation

In order to run the batch of experiments, a new set of instances was created for the multi-objective
formulation RM |Si jk |(Cmax , TEC). This set was inspired by the instances proposed by Vallada and
Ruiz (2011) and available at SOA (2011). The new set of instance has 80 problems, considering
combinations of 6, 8, 10, 12 and 15 jobs with 2, 3, 4 and 5 machines. The combinations of 6, 8, 10
and 12 jobs have 3 modes of operation (q = 3), representing normal speed, slow and fast, while the
combinations of 15 jobs have five modes of operation (q = 5). Table 2 summarizes the characteristics
of the test instances and the source of the original instance.

5.2. Method used for solving the proposed mathematical model

The -constrained method and the smart pool (described in Section 4) were used for solving the
proposed model. In the -constrained method, one of the objective functions is kept and the other
ones are placed as inequality constraints. Thus, the -constrained method converts a multi-objective
problem into a single-objective one with additional constraints. The following lines (13) and (14)
present an example of the method for a minimization problem using the -constrained strategy.
min f1 (x) (13)
Subject to:
fi (x) ≤ i , i = 2, . . . , r. (14)

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 13
Table 3
Average time to solve single-objective versions of the problem (-constrained method)

Instance set Minimizing the Cmax Minimizing the TEC

6 0.113 0.333
8 0.313 1.360
10 1.082 42.744
12 6.989 556.885
15 224.263 2402.626

Table 4
Smart pool proposed variants

Number of MILP
Acronym wmax timeLim (seconds) problems
SmartPool1 10 20 10
SmartPool2 20 20 20

In this problem, x is the decision variable and there is a total of r objectives. The objective f1 is
defined as the most important, being considered the only objective in this formulation. The other
objectives fi (with i = 2, . . . , r) are converted into constraints. The parameter i is the upper bound
of objective fi . Different solutions on the Pareto-optimal front can be achieved upon variations of
i .
For application in this work, initially we generate for each instance one single-objective mathemat-
ical model minimizing only the Cmax and another single-objective mathematical model minimizing
only the TEC. In each case, the other objective was kept as a constraint. A similar procedure was
done in Toro et al. (2017), in the context of multi-objective vehicle routing problems, in order to
identify which objective should be selected in the -constrained method. Table 3 presents the average
time (in seconds) for the executions of the single-objective mathematical models. The best results
are highlighted in bold. These values are grouped by instances with the same number of jobs.
It can be observed that the model minimizing the Cmax is much more efficient. Thereby, in the
-constraint method the objective of minimizing makespan (Cmax ) is defined as the most important
goal and the TEC is considered as a constraint.
The implementation of the -constrained method is detailed next. Initially, the maximum point fb
was defined, as well as the minimum point fa for objective TEC, for each instance. For calculating fa ,
the single-objective mathematical model minimizing only the TEC was used. Similarly, to calculate
the fb, the single-objective mathematical model minimizing only the Cmax was used. Thus, it was
possible to calculate the corresponding values of the TEC. Ten different values for  were created
by an uniform distribution (U [ fa , fb]) for each test problem.
For the smart pool method, two different configurations were defined (SmartPool1 and
SmartPool2 ), ensuring the method to solve approximately the same amount of problems generated
by the -constrained method. The configuration SmartPool1 solves 10 problems and the config-
uration SmartPool2 solves 20 problems. Each problem could be optimized with a maximum time
limit of 20 seconds. This time limit was obtained previously by empirical tests. We have obtained a

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
14 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
300

280

260

240

220
Makespan

200

180

160

140

120

100
200 300 400 500 600 700 800
TEC

Fig. 3. Example 1: Pareto front for a given test instance taken at random.
[Colour figure can be viewed at wileyonlinelibrary.com]

minimum time limit to solve each problem satisfactorily. Table 4 shows the characteristics of these
configurations.

5.3. Analysing the trade-off between both objectives

Next, the estimates of the Pareto front found by the -constrained method, considering two instances
taken at random, are presented. These sets of non-dominated solutions are shown in Figs. 3 and
4. Blue points indicate the Pareto solutions for the RM |Si jk |(Cmax , TEC) and red points show
the optimal solution for the RM |Si jk |Cmax , considering the makespan as a single objective to be
minimized.
It is observed that there is great variability in the values of the two goals. The point in red
colour is important for highlighting that only this solution would be analysed in a single-objective
version of the problem. As depicted in Fig. 3, for example, about 700 units of energy would be
required to implement the solution that minimizes the makespan (hence maximizing production).
These graphics illustrate that both objectives (Cmax and TEC) are conflicting and emphasize the
importance of this multi-objective approach for the decision maker. By selecting a different solution
on the Pareto front, the decision maker could choose to sacrifice some units of makespan in order
to achieve less energy consumption.
To better illustrate the conflict between both objectives, Table 5 shows the maximum and minimum
values found, for each objective, by the -constrained method. These values are grouped by instances

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 15
280

260

240

220
Makespan

200

180

160

140

120
450 500 550 600 650 700 750 800 850 900
TEC

Fig. 4. Example 2: Pareto front for a given test instance taken at random.
[Colour figure can be viewed at wileyonlinelibrary.com]

Table 5
Maximum and minimum values of the solutions found by the -constrained method

Cmax TEC
Instance set Max Min Difference Max Min Difference
6 383.0 23.0 360.0 713.0 57.0 656.0
8 359.0 36.0 323.0 1223.0 146.0 1077.0
10 524.0 55.0 469.0 1125.0 212.0 913.0
12 634.0 59.0 575.0 1738.0 173.0 1565.0
15 469.0 54.0 415.0 1649.0 220.0 1429.0

with the same number of jobs. It can be verified that there is a great difference between the maximum
and minimum values of both goals. In the sets involving 12 jobs, there is a difference of 1565 units
regarding total electricity consumption, showing the importance of this goal for the problem.

5.4. Comparing results between -constrained method and smart pool

Table 6 shows the average time taken for finding the non-dominated solutions in the first batch of
experiment of the -constrained method. The average time is given by the time spent (in seconds)
divided by the number of non-dominated solutions found. These values are grouped by the instances
with same number of jobs.

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
16 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
Table 6
Average time for finding non-dominated solutions – -constrained method and smart pool configurations

Instance set -constrained SmartPool1 SmartPool2

6 0.208 (± 0.073) 0.079 (± 0.026 ) 0.150 (± 0.058)
8 0.644 (± 0.289) 0.312 (± 0.329) 0.779 (± 0.468)
10 1.780 (± 1.421) 0.807 (± 0.231) 1.277 (± 0.459)
12 3.994 (± 3.781) 1.371 (± 0.712) 1.953 (± 1.063)
15 54.663 (± 88) 6.998 (± 5.104) 8.172 (± 5.009)

Table 7
Results of indicator HV

Instance set -constrained SmartPool1 SmartPool2

6 0.708 (± 0.086) 0.810 (± 0.070) 0.821 (± 0.061)
8 0.768 (± 0.093) 0.814 (± 0.118) 0.826 (± 0.116)
10 0.791 (± 0.055) 0.845 (± 0.078) 0.858 (± 0.064)
12 0.863 (± 0.093) 0.934 (± 0.072) 0.942 (± 0.068)
15 0.833 (± 0.088) 0.884 (± 0.113) 0.909 (± 0.087)

The best results are highlighted in bold. It is observed that the SmartPool1 method achieved the
best results in all cases. It can also be verified that the -constrained method took a lot of time to
solve the instances with 15 jobs.
In order to check the quality of the obtained sets of non-dominated solutions, two indicators of
quality were considered: the hyper-volume (HV) and the coverage between two sets (CS) (Zitzler
and Thiele, 1999). HV indicator of a set P calculates the volume of the region between the points
pi ∈ P and a reference point rp. For each solution pi ∈ P, a hyper-cube hyi is constructed according
to the reference point rp. In maximization problems, the reference point usually is (0,0), while in
minimization ones, the upper limit should be determined for each objective. The HV of an estimate
of the Pareto front is the sum of hyper-cubes each solution set contains. In this current work, the
reference point rp is the worst solution for both objectives, which are calculated in the same way
that these values are used for calculating the  values.
The CS indicator determines the percentage of solutions of another set (X ) that a particular set
(X ) dominates. Equation (15) shows the calculation of the coverage indicator:

|a ∈ X ; ∃a ∈ X : a covers a |
CS(X , X ) = . (15)
|X |
Operation a covers a determines that a dominates a or a is equal to a . The results of CS
indicator are mapped from [0,1]. If the result of CS is equal to 1 implies that all points X are
dominated or equal to those in X .
Tables 7 and 8 show the obtained results for both indicators. Values are also reported according
to groups of instances with the same number of jobs.
The best results are highlighted in bold. For the HV metric, SmartPool2 method achieved the
best results in all cases; however, the values are very similar to each other. This indicates that all

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 17
Table 8
Results for indicator CS

-constrained × SmartPool1 × -constrained × SmartPool2 ×

Instance set SmartPool1 -constrained SmartPool2 -constrained
6 0.067 (± 0.129) 0.111 (± 0.136) 0.043 (± 0.0701) 0.132 (± 0.13)
8 0.097 (± 0.096) 0.116 (± 0.098) 0.057 (± 0.0978) 0.116 (± 0.093)
10 0.063 (± 0.054) 0.092 (± 0.122) 0.053 (± 0.0611) 0.090 (± 0.112)
12 0.062 (± 0.065) 0.053 (± 0.069) 0.061 (± 0.0384) 0.076 (± 0.096)
15 0.125 (± 0.088) 0.027 (± 0.057) 0.084 (± 0.0638) 0.045 (± 0.081)
0.9
0.8
0.7
HV

0.6
0.5
0.4

Epsilon...constrained SmartPool1 SmartPool2

Métodos

Fig. 5. Box plot of results for the HV indicator.

methods presented good convergence to the true Pareto front of the problems. This is expected for
the -constrained method, since it relies on the exact solution provided by the solver. In smart pool
method, the solver does not necessarily run until optimality, because of the time limit parameter.
Nevertheless, for the instances considered, the convergence is quite good provided the values of
HV metric. Regarding the CS indicator, the smart pool method also got the best results in most
cases with the two configurations. But it was observed that both methods achieved low coverage
relative to each other, thus suggesting that both methods match high-quality solutions relative to
their estimated Pareto sets. Figures 5 and 6 show box plot graphs.
Figure 7 provides a graphical analysis of the -constrained method and smart pool, using two
instances randomly selected. One instance has 12 jobs and 3 machines and the other has 15 jobs
and 2 machines. A new variant of smart pool was created, the SmartPool3 , with wmax = 100 and
timeLim = 600 (seconds). This proposed combination of parameters may be able to show the
evolution of the method with more weights and long time limit running for each solved MILP.
Figures 7 and 8 show the results of the methods. The results of -constrained method are optimal.
It can be seen that the smart pool search was able to find a very good approximation of the Pareto
front with a much larger number of solutions.

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
18 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22

0.5
0.4
0.3
CS

0.2
0.1
0.0

Epsilon...const.x.SmartPool1 Epsilon...const.x.SmartPool2 SmartPool1.x.Epsilon...const SmartPool2.x.Epsilon...const

Methods

Fig. 6. Box plot of results for the CS indicator.

800

600
Makespan

Methods
Epsilon−constrained
SmartPool3

400

200

400 500 600 700 800 900

TEC

Fig. 7. Obtained Pareto fronts for the -constrained method and SmartPool3 for an instance with 12 jobs and 3
machines. [Colour figure can be viewed at wileyonlinelibrary.com]

6. Conclusions

This work presented a multi-objective approach for the unrelated parallel machine scheduling
problem with sequence-dependent setup times, in the context of green scheduling, in which the
main goals were to minimize makespan and TEC, formally defined as RM |Si jk |(Cmax , TEC).
Motivated by the lack of similar approaches in the literature, we highlighted the possibility of
providing good-quality schedules with low consumption of energy. For this purpose, a bi-objective
variant of the problem was designed with a mathematical model. This model was inspired by recent

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 19

275

250
Makespan

Methods
225 Epsilon−constrained
SmartPool3

200

175

700 800 900 1000

TEC

Fig. 8. Obtained Pareto fronts for the -constrained method and SmartPool3 for an instance with 15 jobs and 2
machines. [Colour figure can be viewed at wileyonlinelibrary.com]

works from the literature, which are moving towards the consideration of energy consumption in
scheduling problems. The proposed mathematical model employs additional features that made the
analysis of machines energy consumption possible, such as variable speed options for processing a
job and corresponding machine power inputs.
For conducting our experiments, a new set of instances was created. The use of the multi-
objective smart pool search matheuristic (or smart pool) was considered in order to find sets of
non-dominated solutions in fixed and restricted computational budget, and compared with the
classical -constrained method. The trade-off between both objectives, makespan and TEC, was
verified and analysed. It was noted that the average time, per non-dominated solution found, taken
by the smart pool was lower than the -constrained method, mainly for the larger problem instances
(with 15 jobs). Thus, in real-life applications, this method would give more possibilities for decision
makers in industry. The HV and coverage indicators also suggests a better performance of the smart
pool. But it was observed that both methods achieved low coverage relative to each other.
In general, the contributions of this work can be applied to others scheduling problems. The
mathematical model proposed and the smart pool search matheuristic can be easily adapted to
various other machine scheduling problems. In this work, it is shown that the objectives makespan
and TEC are conflicting; and that the TEC is very important because it represents a big cost to the
industry. The efficiency of the smart pool search also is revealed, showing that this matheuristic can
be applied to countless other problems in a simple way.
As future work, we intend to expand the proposed set of instances, as well as develop metaheuristic
algorithms for solving the RM |Si jk |(Cmax , TEC). Given the complexity of the problem, its use might
be essential for solving large instances.

C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
20 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
Acknowledgements

The authors would like to thank the Brazilian funding agencies CAPES, CNPq (grants 306850/2016-
8, 307915/2016-6) and FAPEMIG (grant PPM CEX 676/17). Vitor N. Coelho would like to thank
FAPERJ (grant E-26/202.868/2016).

References

Agnetis, A., Detti, P., Meloni, C., Pacciarelli, D., 2001. Set-up coordination between two stages of a supply chain. Annals
of Operations Research 107, 15–32.
Ahilan, C., Kumanan, S., Sivakumaran, N., Dhas, J.E.R., 2013. Modeling and prediction of machining quality in CNC
turning process using intelligent hybrid decision making tools. Applied Soft Computing 13, 3, 1543–1551.
Al-Salem, A., 2004. Scheduling to minimize makespan on unrelated parallel machines with sequence dependent setup
times. Engineering Journal of the University of Qatar 17, 1, 177–187.
Arnaout, J.P., Musa, R., Rabadi, G., 2014. A two-stage ant colony optimization algorithm to minimize the makespan on
unrelated parallel machines – part II: enhancements and experimentations. Journal of Intelligent Manufacturing 25,
43–53.
Arnaout, J. P., Rabadi, G., Musa, R., 2010. A two-stage ant colony optimization algorithm to minimize the makespan
on unrelated parallel machines with sequence-dependent setup times. Journal of Intelligent Manufacturing 21, 6,
693–701.
Avalos-Rosales, O., Alvarez, A., Angel-Bello, F., 2013. A reformulation for the problem of scheduling unrelated par-
allel machines with sequence and machine dependent setup times. Proceedings of the Twenty-Third International
Conference on Automated Planning and Scheduling, Rome, Italy, pp. 278–282.
Avalos-Rosales, O., Angel-Bello, F., Alvarez, A., 2015. Efficient metaheuristic algorithm and re-formulations for the
unrelated parallel machine scheduling problem with sequence and machine-dependent setup times. International
Journal of Advanced Manufacturing Technology 76, 9, 1705–1718.
Bampis, E., Letsios, D., Lucarelli, G., 2015. Green scheduling, flows and matchings. Theoretical Computer Science 579,
126–136.
Cheng, J., Chu, F., Liu, M., Wu, P., Xia, W., 2017. Bi-criteria single-machine batch scheduling with machine on/off
switching under time-of-use tariffs. Computers & Industrial Engineering 112, 721–734.
Coelho, B.N., Coelho, V.N., Coelho, I.M., Ochi, L.S., Roozbeh Haghnazar, K., Lima, M.S., da Costa, A.R., 2017a. A
multi-objective green UAV routing problem. Computers & Operations Research 88, Supplement C, 306–315.
Coelho, I.M., Munhoz, P.L.A., Haddad, M.N., Coelho, V.N., Silva, M.M., Souza, M.J.F., Ochi, L.S., 2011. A computa-
tional framework for combinatorial optimization problems. VII ALIO/EURO Workshop on Applied Combinatorial
Optimization, Porto, pp. 51–54.
Coelho, V.N., Coelho, I.M., Coelho, B.N., Cohen, M.W., Reis, A.J.R., Silva, S.M., Souza, M.J.F., Fleming, P.J., Guimarães,
F.G., 2016a. Multi-objective energy storage power dispatching using plug-in vehicles in a smart-microgrid. Renewable
Energy 89, 730–742.
Coelho, V.N., Coelho, I.M., Coelho, B.N., de Oliveira, G.C., Barbosa, A.C., Pereira, L., de Freitas, A., Santos, H.G.,
Ochi, L.S., Guimarães, F.G., 2016b. A communitarian microgrid storage planning system inside the scope of a smart
city. Applied Energy 201, 371–381.
Coelho, V.N., Cohen, M.W., Coelho, I.M., Liu, N., Guimarães, F.G., 2017b. Multi-agent systems applied for energy
systems integration: state-of-the-art applications and trends in microgrids. Applied Energy 187, 820–832.
Cota, L.P., Guimarães, F.G., Oliveira, F.B., Souza, M.J.F., 2017. An adaptive large neighborhood search with learn-
ing automata for the unrelated parallel machine scheduling problem. Proceedings of the 2017 IEEE Congress on
Evolutionary Computation (CEC 2017), Donostia, San Sebastián, pp. 185–192.
Cota, L.P., Haddad, M.N., Souza, M.J.F., Coelho, V.N., 2014a. AIRP: a heuristic algorithm for solving the unrelated
parallel machine scheduling problem. Proceedings of the 2014 IEEE Congress on Evolutionary Computation (CEC
2014), Beijing, pp. 1855–1862.


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22 21
Cota, L.P., Haddad, M.N., Souza, M.J.F., Martins, A.X., 2014b. A heuristic algorithm for solving the unre-
lated parallel machine scheduling problem with sequence dependent setup time (in Portuguese). Proceed-
ings of the 35th Brazilian Congress of Applied and Computational Mathematics, Natal, Brazil. Available at
http://proceedings.sbmac.org.br/sbmac/article/view/724/730 (accessed January 2017).
Deb, K., Jain, H., 2014. An evolutionary many-objective optimization algorithm using reference-point-based nondomi-
nated sorting approach, Part I: solving problems with box constraints. IEEE Transactions on Evolutionary Computa-
tion 18, 4, 577–601.
Deb, K., Pratap, A., Agarwal, S., Meyarivan, T., 2002. A fast and elitist multiobjective genetic algorithm: NSGA-II.
IEEE Transactions on Evolutionary Computation 6, 182–197.
Ding, J., Song, S., Wu, C., 2016. Carbon-efficient scheduling of flow shops by multi-objective optimization. European
Journal of Operational Research 248, 758–771.
Franca, P.M., Gendreau, M., Laporte, G., Muller, F.M., 1996. A tabu search heuristic for the multiprocessor scheduling
problem with sequence dependent setup times. International Journal of Production Economics 43, 79–89.
Garey, M., Johnson, D., 1979. Computers and Intractability: A Guide to the Theory of NP-Completeness. W.H. Freeman
& Co., San Francisco, CA.
Glass, A.C., Potts, C.N., Shade, P., 1994. Unrelated parallel machine scheduling using local search. Mathematical and
Computer Modeling 20, 41–52.
Graham, R., Lawler, E., Lenstra, J., Kan, A., 1979. Optimization and approximation in deterministic sequencing and
scheduling: a survey. Annals of Discrete Mathematics 5, 2, 287–326.
Haddad, M.N., Cota, L.P., Souza, M.J.F., Maculan, N., 2014. AIV: a heuristic algorithm based on iterated local search
and variable neighborhood descent for solving the unrelated parallel machine scheduling problem with setup times.
Proceedings of the 16th International Conference on Enterprise Information Systems (ICEIS 2014), Lisbon, Portugal,
pp. 376–383. https://doi.org/10.5220/0004884603760383 (accessed January 2017).
Haddad, M.N., Cota, L.P., Souza, M.J.F., Maculan, N., 2015. Solving the unrelated parallel machine scheduling problem
with setup times by efficient algorithms based on iterated local search. Lecture Notes in Enterprise Information
Systems 227, 131–148.
Hansen, P., Mladenovic, N., Pérez, J.A.M., 2008. Variable neighborhood search: methods and applications. 4OR: Quarterly
Journal of the Belgian, French and Italian Operations Research Societies 6, 319–360.
Karp, R.M., 1972. Reducibility among combinatorial problems. Complexity of Computer Computations 40, 4, 85–103.
Kim, D.W., Kim, K.H., Jang, W., Chen, F.F., 2002. Unrelated parallel machine scheduling with setup times using simulated
annealing. Robotics and Computer-Integrated Manufacturing 18, 223–231.
Kim, D.W., Na, D.G., Chen, F.F., 2003. Unrelated parallel machine scheduling with setup times and a total weighted
tardiness objective. Robotics and Computer-Integrated Manufacturing 19, 173–181.
Kurz, M.E., Askin, R.G., 2001. Heuristic scheduling of parallel machines with sequence-dependent set-up times. Inter-
national Journal of Production Research 39, 3747–3769.
Lin, S.W., Ying, K.C., 2014. ABC-based manufacturing scheduling for unrelated parallel machines with machine-
dependent and job sequence-dependent setup times. Computers & Operations Research 51, 172–181.
Liu, Y., Dong, H., Lohse, N., Petrovic, S., 2016. A multi-objective genetic algorithm for optimisation of energy consump-
tion and shop floor production performance. International Journal of Production Economics 179, 259–272.
Logendran, R., McDonell, B., Smucker, B., 2007. Scheduling unrelated parallel machines with sequence-dependent
setups. Computers & Operations Research 34, 11, 3420–3438.
Lourenço, H.R., Martin, O., Stützle, T., 2003. Iterated local search. In Glover, F., Kochenberger, G. (eds), Handbook
of Metaheuristics, International Series in Operations Research & Management Science, Vol. 57. Kluwer Academic,
Norwell, MA, pp. 321–353.
Lust, T., Teghem, J., 2010. Two-phase pareto local search for the biobjective traveling salesman problem. Journal of
Heuristics 16, 475–510.
Mansouri, S.A., 2005. Coordination of set-ups between two stages of a supply chain using multi-objective genetic
algorithms. International Journal of Production Research 43, 3163–3180.
Mansouri, S.A., Aktas, E., 2016. Minimizing energy consumption and makespan in a two-machine flow shop scheduling
problem. Journal of the Operational Research Society 67, 11, 1382–1394.


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies
22 L. P. Cota et al. / Intl. Trans. in Op. Res. 00 (2018) 1–22
Mansouri, S.A., Aktas, E., Besikci, U., 2016. Green scheduling of a two-machine flow shop: trade-off between makespan
and energy consumption. European Journal of Operation Research 248, 772–788.
Meneghini, I.R., Guimarães, F.G., 2017. Evolutionary method for weight vector generation in multi-objective evolutionary
algorithms based on decomposition and aggregation. 2017 IEEE Congress on Evolutionary Computation (CEC),
San Sebastian, Spain, pp. 1900–1907.
Pereira Lopes, M.J., de Carvalho, J.M., 2007. A branch-and-price algorithm for scheduling parallel machines with
sequence dependent setup times. European Journal of Operational Research 176, 1508–1527.
Pinedo, M.L., 2008. Scheduling: Theory, Algorithms, and Systems (3rd edn). Springer, New York, NY.
Rabadi, G., Moraga, R.J., Al-Salem, A., 2006. Heuristics for the unrelated parallel machine scheduling problem with
setup times. Journal of Intelligent Manufacturing 17, 1, 85–97.
Radhakrishnan, S., Ventura, J.A., 2000. Simulated annealing for parallel machine scheduling with earliness/tardiness
penalties and sequence-dependent set-up times. International Journal of Production Research 38, 2233–2252.
Randhawa, S.U., Kuo, C.H., 1997. Evaluating scheduling heuristics for non-identical parallel processors. International
Journal of Production Research 35, 969–981.
Santos, H.G., Toffolo, T.A., Silva, C.L., Vanden Berghe, G., 2016. Analysis of stochastic local search meth-
ods for the unrelated parallel machine scheduling problem. International Transactions in Operational Research.
https://doi.org/10.1111/itor.12316.
Sauer, I.L., Tatizawa, H., Salotti, F.A., Mercedes, S.S., 2015. A comparative assessment of Brazilian electric motors
performance with minimum efficiency standards. Renewable and Sustainable Energy Reviews 41, Suppl. C, 308–318.
Scheffé, H., 1958. Experiments with mixtures. Journal of the Royal Statistical Society 20, 2, 344–360.
Sen, P.C., 2013. Principles of Electric Machines and Power Electronics. Wiley, Hoboken, NJ.
SOA, 2011. Sistemas de Optimizacion Aplicada. A web site that includes benchmark problem data sets and solutions for
scheduling problems. Available at http://soa.iti.es/problem-instances (accessed January 2017).
Srivastava, B., 1997. An effective heuristic for minimizing makespan on unrelated parallel machines. Journal of the
Operational Research Society 49, 886–894.
Toro, E.M., Franco, J.F., Echeverri, M.G., Guimarães, F.G., 2017. A multi-objective model for the green capacitated
location-routing problem considering environmental impact. Computers & Industrial Engineering 110, 114–125.
Tran, T.T., Araujo, A., Beck, J.C., 2016. Decomposition methods for the parallel machine scheduling problem with setups.
INFORMS Journal on Computing 28, 1, 83–95.
Vallada, E., Ruiz, R., 2011. A genetic algorithm for the unrelated parallel machine scheduling problem with sequence
dependent setup times. European Journal of Operational Research 211, 3, 612–622.
Wang, S., Liu, M., Chu, F., Chu, C., 2016. Bi-objective optimization of a single machine batch scheduling problem with
energy cost consideration. Journal of Cleaner Production 137, 1205–1215.
Weng, M.X., Lu, J., Ren, H., 2001. Unrelated parallel machine scheduling with setup consideration and a total weighted
completion time objective. International Journal of Production Economics 70, 215–226.
Wildi, T., 2013. Electrical Machines, Drives and Power Systems. Pearson, New York.
Ying, K.C., Lee, Z.J., Lin, S.W., 2012. Makespan minimisation for scheduling unrelated parallel machines with setup
times. Journal of Intelligent Manufacturing 23, 5, 1795–1803.
Yu, J.J., 2010. Green scheduling and its solution. In Zhang, L., Zhang, C., Shi, T. (eds) Advanced Materials Research, Vol.
139–142. Trans Tech Publications, Zurich, pp. 1415–1418.
Zhang, H., Zhao, F., Fang, K., Sutherland, J.W., 2014. Energy-conscious flow shop scheduling under time-of-use electricity
tariffs. CIRP Annals 63, 37–40.
Zhou, A., Qu, B.Y., Li, H., Zhao, S.Z., Suganthan, P.N., Zhang, Q., 2011. Multiobjective evolutionary algorithms: a
survey of the state of the art. Swarm and Evolutionary Computation 1, 1, 32–49.
Zhu, X., Wilhelm, W.E., 2006. Scheduling and lot sizing with sequence-dependent setup: a literature review. IIE Transac-
tions 38, 987–1007.
Zitzler, E., Thiele, L., 1999. Multiobjective evolutionary algorithms: a comparative case study and the strength Pareto
approach. IEEE Transactions on Evolutionary Computation 3, 257–271.


C 2018 The Authors.
International Transactions in Operational Research 
C 2018 International Federation of Operational Research Societies