Applied Energy 249 (2019) 300–315

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

The impact of various carbon reduction policies on green flowshop T

scheduling
Mehdi Foumani∗, Kate Smith-Miles
School of Mathematics and Statistics, The University of Melbourne, Parkville 3010, Australia

HIGHLIGHTS

• Presenting a set of mathematical models for green flowshop scheduling problems.

• Considering common carbon reduction policies and their corresponding scenarios.
• Providing analysis of models and insights to design a framework for green flowshops.
• Industries can use our cost-effective strategies to meet limits mandated by government.
• Government can explore our integrated framework to gauge impacts of carbon policies.

ARTICLE INFO ABSTRACT

Keywords: In this paper we consider, from an environmental policy-maker perspective, how carbon reduction policies
Green flowshop scheduling impact the economic competitiveness of the manufacturing sector. Specifically, we focus on flowshop scheduling
Carbon reduction policy – which typically aims to minimize makespan for purely economic objectives – and consider how three common
Energy consumption carbon reduction policies – namely, taxes on emissions, baselines on emissions, and emissions trading schemes –
Mixed-integer linear programming
can create competitive green flowshops that balance minimization of makespan and carbon emissions. The goal is
Bi-objective optimization
to enable policy-makers to understand how to set policies and control parameters to achieve environmental
Sensitivity analysis
objectives while ensuring global economic competitiveness of industry. We initially present a set of mixed-
integer linear programming (MILP) models for flowshop scheduling problems operating in a regulated en-
vironment in which each carbon reduction policy is adopted. We then introduce a bi-objective scheduling fra-
mework for the corresponding problem to obtain alternative solutions under each policy. These models and their
computational results however, are not the main focus of the study, but are presented as a means to demonstrate
how green policies co-exist with economic objectives, with policy-makers in control of the balance. To this end,
based on financial data from Australia’s carbon emissions profile, we provide a policy-oriented analysis of the
models, and some managerial insights into the effect of scheduling strategies on carbon emissions under different
reduction policies. These insights offer support to both environmental policy-makers and corporate production
and sustainability managers to determine whether it is technically feasible and profitable to replace traditional
scheduling strategies with environmentally friendly scheduling strategies.

1. Introduction policy actions at the individual, organizational, or program level. Such

studies have so far focused on areas including supply chain manage-
Due to the continuing depletion of environmental resources, the ment [1], cloud computing [2], power systems [3], life cycle analysis
conflict between economic growth and environmental protection has [4], recycling [5], and manufacturing operations scheduling [6].
justifiably received global attention. Central to the challenge is a critical Of increasing importance for global manufacturing sectors is the
policy-maker problem: how to design policies such that an appropriate sustainable scheduling of operations and processes. Despite differences
balance is achieved between global economic competitiveness of in- in economic development level, energy consumption structure and
dustry and incentives to reduce resource consumption and environ- differences in local industry, the push for reduced carbon emissions is
mental harm. Policy-oriented research studies evaluate the effects of becoming a priority worldwide [7]. In China, for example, the

∗
Corresponding author.
E-mail addresses: [email protected] (M. Foumani), [email protected] (K. Smith-Miles).

https://doi.org/10.1016/j.apenergy.2019.04.155
Received 11 March 2019; Received in revised form 23 April 2019; Accepted 28 April 2019
Available online 06 May 2019
0306-2619/ © 2019 Elsevier Ltd. All rights reserved.
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

industrial sectors are the source of almost 50% of carbon emissions [8], • Finally, from a tactical point of view, governmental policy-makers
and consequently carbon trading pilot programs are being explored in can use our integrated framework to gauge the relative impact of
China [9,10]. In line with global sustainability efforts, the Australian carbon policies as a driver of corporate green actions.
government has regulated industrial sectors to adopt energy-saving
mechanisms, or equivalent carbon reduction policies. An Australian The first contribution views the problem as a shop scheduling pro-
carbon reduction policy, known as the safeguard mechanism, was re- blem, but the last two contributions consider the problem as a policy
leased in July 2016. The safeguard mechanism covers a wide variety of making problem. Although obtained results are interpreted in terms of
industry sectors such as electricity generation, mining, oil and gas, both shop scheduling and energy policy making, the last two con-
transport, construction, waste and manufacturing. Collectively, it tributions play a more important role in this study. This is to say, the
covers almost half of Australia’s carbon emissions [11]. It is worthwhile structural results focus mainly on environmental and energy themed
to note that the goals of the safeguard mechanism are reached through contributions that are neglected in the literature, rather than im-
establishing taxes on emissions, baselines on emissions and effective provements to mathematical programming techniques for flowshop
trading of emissions. Accordingly, large emitters are required to either scheduling problems that have already received a significant amount of
pay an emissions tax for every tonne of carbon released, maintain attention in the literature. The key analysis tool utilized to achieve
emissions below an agreed baseline, or to purchase Australian Carbon these aims is sensitivity analysis to identify the critical control para-
Credit Units to offset emissions which exceed their baseline. This leads meters for the design of an optimized system, as this method is em-
to the following key question: Given the safeguard mechanism involves ployed extensively in the field of energy/CO2 emission analysis
regulatory intrusion by mandating individual carbon emissions, how can [16–19].
other corporate priorities, such as profitability, be controlled to ensure global In keeping with GFSPs, the following definitions are used to de-
competitiveness? The effective implementation of any government scribe the main carbon reduction policies referred throughout the
policy-driven safeguard mechanism lies in its ability to answer this paper:
question about corporate competitiveness, and is the focus of this study.
Since the flowshop is one of the most common process models in • Tax on Emissions (ToE) carbon reduction policy: GFSPs are taxed by
many manufacturing [12] and services industries and it is the most government on the amount of carbon emitted throughout the ma-
popular representative of series production systems, this paper offers a kespan.
pilot study that shows how the right policies can enable manufacturers • Baseline on Emissions (BoE) carbon reduction policy: GFSPs are
to use existing solution methods for solving flowshop scheduling pro- subject to mandatory baselines (upper bounds) on the amount of
blems (FSP) to produce schedules that are economically viable and carbon emitted throughout the makespan.
environmentally friendly. The analysis provides some insights into the • Emissions Trading (ET) carbon reduction policy: although GFSPs are
behaviour of flowshop process models in the presence of environmental still subject to carbon baselines, they can either receive a reward or
protection regulations, and how control parameters can be set to penalty for emitting carbon under or over the baselines throughout
achieve desired outcomes. Classical FSPs, which do not address carbon the makespan, respectively.
emission concerns, have been well studied in the literature for several
decades, and we refer the interested readers to a recent book [13] for a Note that the third policy is effectively a combination of the first
comprehensive discussion of the many variants for different shop en- two policies in that both taxes and baselines are included in this policy.
vironments. The computational study of variations of FSP is also well It has the potential to capture the advantages of both the ToE and BoE
studied, especially for bicriteria FSPs [14]. Our study is focused on a policies. In order to demonstrate this, we study the performance of all
particular extension of the classical model, known as a green flowshop, three policies and their ability to support the goals of decision makers
in which carbon emissions are a key concern. We emphasize here that through GFSP scheduling decisions.
this is not a computational study on improved formulations and algo- The overall structure of this paper is as follows. In Section 2, we
rithms for solving FSPs with multiple objectives related to economic describe our research problem, and then discuss the relevant literature
and environmental concerns. Rather than a computational study of how on GFSPs. In Section 3, we formulate MILP models for various carbon
best to solve a model with environmental constraints or objectives, our reduction policies. Section 4 is the core of the paper where we conduct
focus here is to explore how different carbon reduction policies can a sensitivity analysis, using newly generated benchmark instances, to
impact the quality of scheduling solutions found by a typical solver provide a set of managerial insights about the effect of carbon reduction
used by industry (CPLEX), and to explore how the control parameters of policies on emissions from optimized schedules. Finally, Section 5
the policy affect the competitiveness of the solutions. This policy-or- concludes the paper with perspectives on future opportunities on this
iented perspective will be critical to enabling any solver technology, important topic.
regardless of the algorithm, to deliver benefit to society through in-
forming policy decisions [15]. Thus, this paper initially reviews the
literature associated with green flowshop scheduling problems (GFSPs). 2. Problem description and related literature
We then extend this literature by making the following novel con-
tributions: As input to the FSP, we typically have three components: a set of m
given machines denoted by = {1, 2, …, m} , a set of n jobs
• From a modelling point of view, we improve the existing mixed- = {1, 2, …, n} from the same family, and an m-by-n matrix of the time
integer linear programming (MILP) models of GFSPs in order to taken for each job to be processed by each machine. All jobs and ma-
capture the structure of different carbon reduction policies and their chines are independent and available at time t = 0. Each of the n jobs
corresponding scenarios that have been considered in the literature must be processed on m machines in the same order. A particular job
of GFSPs. We then provide analysis of the models and a set of in- j can start its processing on machine i only when its operation
sights to design an integrated framework for scheduling of green on machine i − 1 is finished and also only when machine i is empty. It
flowshop process models. follows that each job can be processed on a single machine, and each
• More significantly, from an operational point of view, we offer a tool machine can perform a single operation at each particular point in time.
for corporate production and sustainability managers to use our The FSP with these underlying assumptions can be converted into a
integrated framework to fine-tune their specific parameters and GFSP in one of the following two ways.
define cost-effective carbon abatement strategies that simulta-
neously meet limits mandated by government.

301
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Fig. 1. A schedule and its carbon emission of a GFSP with four machines and three jobs.

2.1. Green variables for FSPs Note that each machine used in FSPs often consumes electricity to
process a single job at each particular point in time since production
For a typical FSP the processing times are predefined, but an im- machines are more related to manufacturing industries such as auto-
portant characteristic of some real world FSPs is that the processing mobile sectors. Hence, here represents a carbon/electricity conver-
time of the part by any particular machine can vary. This characteristic, sion factor. As a brief aside, we note here that it is, of course, possible to
known as controllable processing times, is due to the desirability of consider other cases such as chemical sectors where environmental
varying the processing speeds to control the balance between energy goals may relate more to reduction of wastewater and its toxicity, ra-
usage and job completion times. For FSPs with controllable processing ther than electricity consumption and carbon emissions. Real world
times, the job processing times are often continuous variables that may FSPs reflecting these sectors are named hoist scheduling problems
be reduced in return for extra compression cost, especially in industries (HSPs) in which each machine is able to process a batch of parts si-
with tooling machines and chemical processes [20–23]. Although the multaneously, not necessarily a single part [24]. The electroplating
structure of GFSPs is very similar to FSPs with controllable processing lines are an example of these types of FSPs in practice where the set of
times (in spite of the fact that job processing times is often a discrete machines is replaced with the set of chemical tanks. Consequently, in-
variable in GFSPs), relying on controllable processing times alone is not stead of the carbon/electricity conversion factor, represents the toxic/
sufficient to create a truly green FSP solution without explicitly ad- chemical conversion factor with a different value, as stated in [25].
dressing carbon emission concerns. Suppose we mainly focus on the environmental issues. Then, the
Since carbon emissions in the manufacturing and services industries method for tackling such a FSP related to chemical sectors is almost
are directly related to energy consumption, an effective way to reduce same as FSPs in the automobile sectors in that we only need to consider
carbon emissions is to develop scheduling strategies with the aim of a different value for . However, other scheduling issues (e.g., replacing
reducing energy consumption in the flowshop process model. machines with tanks) may still play a role. The environmentally con-
Accordingly, considering carbon emissions, the main focus of our at- scious HSP is outside the scope of this study, and therefore we refer
tention is on a GFSP with the following common assumptions. Firstly, reader to [25] for a detail study in this direction.
there is a set of s processing speed factors = { 1, 2, …, s } in an in- To help clarify how carbon emissions are estimated in a GFSP, a
creasing order for each machine as the fourth component of the GFSP. schematic picture of a given schedule of a GFSP with four machines and
Note that we are not able to change the processing speed of a machine three jobs is presented in Fig. 1. The top side of Fig. 1 shows the carbon
during execution of a job under this assumption. Each job j takes a pre- emissions for the schedule. For example, carbon emissions at time in-
defined processing time pij on each machine i. However, the associated stance t 0 can be determined as follows: Since the first and third ma-
processing time is pijk =
pij
minutes if the job j is processed on machine i chines are not occupied at t 0 , their carbon emissions are sq1 and sq3 ,
k
respectively. If we assume the second and fourth machines process jobs
at speed factor k . Secondly, each processing speed factor k
3 and 2 with speeds u and v at this instance of time, then
consumes a corresponding amount of energy measured by an energy
the carbon emissions are q23u and q42v , respectively. Accordingly, the
consumption factor k . Therefore, = { 1, 2, …, s } represents the set of
total carbon emissions for the GFSP at t 0 is sq1 + q23u + sq3 + q42v . In
energy consumption factors, which can affect CO2 emissions.
general, we can conclude that the integral of the area under the curve in
Specifically, the associated carbon emissions of pijk is
i pij k
Fig. 1 shows the energy consumption of the entire production period or
qijk = 60 kgCO2 . In this equation, i represents the power of machine equivalently associated carbon emissions for the GFSP. Note that the
k
i in kilowatt (kw) when the machine is in active condition (power is the carbon emissions should be measured by the energy consumption from
rate at which a machine uses energy). Also, represents a conversion time 0 to Cmax since the energy consumption at any time instant, say t 0 ,
factor from the energy consumption to carbon emissions due to a linear is only related to peak energy load in the context of energy demand
relationship between them. Here, a CO2 conversion factor management.
= 0.785 kgCO2 /kw h will be assumed based on an average level, as
stated in Liu et al. [8]. We assume that no machine can be turned off
2.2. Green streams for FSPs
until completion of the last job on the last machine. Alternatively,
considering a stand-by energy consumption factor i , each machine i is
The green variables can be integrated into the FSP in several dif-
running in stand-by mode with carbon emissions sqi = i60i kgCO2 if the
ferent ways. This is not necessarily a trivial task because the form of the
machine is idle. Naturally, i is less than each k to save energy by integration, reflecting the outcome of policy-making, should lead to a
consuming only the power that is needed. good balance between economic and environmental objectives of

302
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

GFSPs. Accordingly, here we provide an overview of existing policies energy load considerations to the problem may end up in a situation
studied and analyzed in the literature. Research on GFSPs has received where the original optimal solution is infeasible. If we want to
significant attention from practitioners and researchers in recent years maintain the makespan, then we have to choose a lower processing
[26], with an assumption about the underlying carbon reduction policy speed for operations (namely type 1 operations) belonging to any
that typically falls into two streams. The first stream of research con- time instant in which the upper bound is violated. Since this cer-
centrates on GFSPs that effectively charge a tax on emissions. The ToE tainly causes a delay in operations afterwards (namely type 2 op-
policy is common in countries such as Denmark, Finland, and Canada erations), we have to choose a higher processing speed for these
[27]. To our best knowledge, there is no previous research in GFSPs operations in order to avoid an increase in makespan. Note that the
that explicitly considers a ToE policy, however, the same outcome has certainty of changing the speed of the type 2 operations comes from
been addressed indirectly by varying energy price based on the time of the optimality of the solution in the sense of energy consumption.
use, as widely addressed in [8,28] for GFSPs. In [29], a NEH-insertion From the original optimal solution, it can be shown that under
algorithm is developed to improve carbon efficiency in GFSPs in which certain conditions the saving in energy consumption of the type 1
machines have variable processing speeds. Numerical results of the operations is less than the increase in energy consumption of the
algorithm show the carbon reduction module of the proposed NEH-in- type 2 operations, meaning that peak energy load and energy con-
sertion algorithm can assist GFSPs in both improving the search effi- sumption are conflicting in nature.
ciency and enhancing the solution quality. Following that, Mansouri • FSPs with only peak energy load [32]: here the situation is similar to
et al. [30] utilized a mathematical model for two-machine GFSPs with Case 2 with the difference that the authors only consider peak en-
sequence dependent setup times. They also developed a heuristic to ergy load. Without loss of generality, considering peak energy load
tackle the combinatorial complexity of the corresponding -hard instead of energy consumption considerations may end up in two
problem. opposite situations: 1. the original optimal solution is infeasible,
The second stream is devoted to GFSPs with BoE policies. Similarly with the logic behind this statement similar to previous case (Case
to the previous stream, strict emission baselines can be reflected by 2). 2. the original optimal solution is still feasible. Then, there is an
energy consumption in the GFSP. The peak energy load and energy opportunity to move away from the original optimal solution and
consumption of machines in a GFSP can be considered as a strict find a solution with shorter makespan without violating the peak
emission baseline under this policy. Therefore, the carbon footprint is energy load. Therefore, a solution with a shorter makespan is im-
related to energy consumption and is obtained by multiplying the plemented although it does not fulfill the need for the reduction of
amount of energy consumed in the flowshop by the carbon/electricity energy consumption and associated carbon emissions. For the sake
emission coefficient, and this value should be less than a given of clarity, the two situations are graphically represented in Figs. 2a
threshold. In [31], a mathematical model is developed for two-machine and 2b for the original optimal solution, and their revised solutions
GFSPs under restriction of peak load and energy consumption. They in Figs. 2c and 2d, respectively.
included the operation speed as an independent variable that can be
changed to affect the peak load and energy consumption. In another Hence, it can be concluded that consideration of terms such as en-
study, Fang et al. [32] extended their model to GFSPs with restrictions ergy saving and environmental-related energy cost is equivalent to con-
on the peak energy load. They also modelled a variant of the problem as sidering energy consumption and associated carbon emissions, and
a traveling salesperson problem (TSP) due to its existing computational these terms can be specified by a set of carbon reduction policies.
benefits via efficient TSP solvers. We refer the interested reader to It can be seen in the studies discussed in this section that a sig-
Gahm et al. [33] for a comprehensive review on GFSPs from various nificant amount of attention is paid in the existing literature to opera-
energy efficiency viewpoints. tional planning of process models. However, it is rare for GFSPs to
A notable feature of carbon emissions is that they are time-in- capture tactical planning decisions in order to set the overall direction
dependent, whereas energy-based objectives are often time-dependent, of organisations. Note that the operational plans are typically short-
as the following three cases from the literature on energy-conscious term in nature and low-level managers are responsible for them,
FSPs illustrate: whereas tactical plans normally cover longer planning horizons with
executive manager and policy-maker responsibility. More specifically,
• FSPs with energy cost and energy consumption [28]: here the au- there is a lack of an integrated framework for GFSPs under different
thors consider energy pricing under time-of-use (TOU) tariffs in carbon reduction policies implemented by governments globally,
FSPs. This is because electricity (or any other type of energy) price especially for an ET carbon reduction policy. Noting this gap that exists
varies hourly to reflect changes in the wholesale electricity market. in the literature, in this paper we study major policies that require
Hence, the energy price can be classified into three levels: on-peak, companies to take actions and manage their carbon emissions. The
mid-peak and off-peak hours. Suppose that we determine the op- mathematical models contribute to the GFSP literature through as-
timal makespan, namely the original optimal solution, for a FSP sisting both manufacturing companies and government regulators to
with only energy consumption considerations (a GFSP). Without loss evaluate the tactical plans under different carbon reduction policies.
of generality, a FSP under TOU tariffs prefers to shift energy usage
from on-peak hours to off-peak or mid-peak hours. Although this
may reduce the energy cost, it increases energy consumption be- 3. Analysis of GFSPs under different carbon reduction policies
cause energy cost and energy consumption are conflicting in nature.
The reason is that we may have to choose higher speed (higher After introducing the general formulation approaches and relevant
energy consumption) for each operation that is shifted from on-peak notations in this section, we present the base case formulation, with no
hours into others in order to maintain the makespan. carbon emissions policy in place. The reason for developing this base
• FSPs with peak energy load and energy consumption [31]: here the case formulation is that the combination of makespan and total carbon
authors consider peak energy load in FSPs, in which violating this emissions leads to a bi-objective GFSP. Hence, in such a case, it makes
peak is forbidden. The peak energy load is the maximum power load sense to model the general case problem first and all other problems
over all time instants, meaning that total power consumption at any (with different carbon reduction policies) afterwards. We will assume
time instant should be less than a fixed upper bound (e.g., the full familiarity with the terminology and notation of bi-objective optimi-
capacity of system). Once again, suppose that we determine the zation (e,g. alternative solutions, scalarization and the -constraint al-
optimal makespan for a FSP with only consideration for energy gorithm), and refer readers to the review paper by Yenisey and
consumption (a GFSP). Without loss of generality, addition of peak Yagmahan [34] for further background on bi-objective FSPs.

303
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Fig. 2. Two situations extracted from consideration of peak energy load instead of energy consumption.

3.1. General formulation approaches It is important to investigate the complexity of the GFSP associated with
carbon emission concerns. Clearly, the two-dimensional matrix of proces-
According to the aforementioned discussion, let us first formulate sing times in the FSP can be converted into a three-dimensional matrix of
the general bi-objective model for the GFSP as follows and label it as P1. processing times in the GFSP where represents the third dimension of
this matrix. This makes it clear that typical FSPs are a special case of GFSPs.
(P1) minimize makespan
Consequently, the GFSP fall within the category of -hard problems.
minimize carbon emissions effects
subject to ordering and timing constraints (1) 3.2. Notation
In model P1, the minimization of makespan achieves economic
benefits by meeting the given demands immediately, and scheduling For the purpose of modelling GFSPs in detail, we adopt the fol-
constraints are ordering and timing constraints similar to classical FSPs. lowing parameters and decision variables throughout the text:
However, the minimization of carbon emissions effects, which origi- Input parameters
nated from the fourth component of GFSPs, distinguishes the analysis of
P1 from other classes of FSPs. m The number of machines
We still need to modify P1 in order to create an interface to the real n The number of jobs
manufacturing and services environment. Therefore, the next major s The number of processing speed factors
step is to bring the dynamic behaviour of P1 into reality by developing pij The processing time of job j on machine i
models P2 and P3 as follows. k The kth processing speed factor of machines (sorted in des-
cending order)
(P2) minimize makespan pijk The processing time of job j on machine i with kth speed
subject to carbon emissions constraints factor
subject to ordering and timing constraints (2) i The power of machine i
k The kth energy consumption factor of machines (sorted in
(P3) minimize carbon emissions effects descending order)
subject to makespan constrains qijk The amount of carbon emitted to process job j on machine i
subject to ordering and timing constraints (3) with speed k
i The stand-by energy consumption factor of machine i
In most cases, decision makers are only interested in minimization sqi The amount of carbon emitted by machine i if it is in a stand-
of makespan or carbon emissions effects, leading to models P2 or P3, by mode.
respectively. Model P2 is more realistic when economic benefits are the The conversion factor from energy to carbon dioxide
primary target of the GFSP, and decision makers are faced with a limit The amount of tax paid on each kilogram of carbon emission
on carbon emissions. Minimization of carbon emissions effects is the The weight parameter combining Cmax and TCE in model P1-
main priority of P3 when the operation is only required to supply ToE
completed jobs before an overall due date, not as soon as possible. It is Q The given baseline of carbon emissions
apparent that the formulations of models P2 or P3 are very similar if The amount of money paid (or received) per unit of carbon
adopting any penalty function approach to handling constraints. credits
Therefore, we limit our analysis to the model P2 in this paper.

304
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

The weight parameter combining Cmax and TCE in model equality of makespan and the completion time of the last process in the
P1P2-ET flowshop. Inequality (7) states that the completion time of the first
M A sufficiently large number associated with the variables process of each job j is greater than or equal to the time taken for
processing the job on the first machine with respect to the selected
processing speed. Likewise, Inequality (8) implies that each job j can
start processing on a machine i only after finishing its operation on the
Outputs upstream machine i 1. As big-M constraints, Inequalities (9) and (10)
jointly ensure the completion time of job j on machine i is shorter than
Cij The completion time for processing job j on machine i in a the completion time of job l on machine i only if job j precedes job l.
given permutation Inequalities (11) and (12) imply precedence relations between any two
Cmax The makespan of a given permutation, i.e., jobs, and also triangle inequality precedence relations between any
Cmax = max {Cij |i , j } three jobs. Eq. (13) states exactly one processing speed should be se-
TCE Total carbon emissions of a given permutation lected from set for processing each job j on machine i, and this
processing speed cannot be changed once the processing of the job is
started. Finally, the restriction on the domain of variables is specified
by Constraint (14). Therefore, the compact form of the model P1 is
Decision variables min {Cmax and TCE|s . t . (6) (14)} . Furthermore, the total number of
variables in the model is n2 + (sm + m 1) n + 2 and the total number
jl A 0/1 variable such that jl = 1 if job j precedes job l, and of constraint is n3 + 2(m 1) n2 + (2m + 1) n .
jl = 0 , otherwise As mentioned earlier, the first objective of P1 is to minimize the
x ijk A 0/1 variable such that x ijk = 1 if machine i processes job j makespan, which is defined as the completion time of the last job on the
with speed k, and x ijk = 0 , otherwise last machine. The makespan reflects the economic concerns of the
Considering the aforementioned parameters and variables, three scheduling strategy since it tends to complete all jobs, or equivalently
different carbon reduction policies and corresponding models of GFSPs meet the given demands, as soon as possible. Therefore, minimization
are detailed in the next subsections. of makespan is often the main goal of production managers. In contrast,
as an environmental concern, the second objective of P1 is to minimize
3.3. Base case formulation: responsiveness trade-off the carbon emissions throughout the manufacturing period. This ob-
jective can inform sustainability managers on how various carbon re-
The base case formulation of GFSPs captures a trade-off balancing duction policies including ToE, BoE, and ET may affect the GFSPs.
makespan and total carbon emissions (with no carbon policy). To demonstrate the conflict between objectives Cmax and TCE, a
Regardless of different carbon reduction policies, let us first provide a GFSP with 10 machines, 10 jobs and 3 processing speed factors is solved
base case formulation of P1, inspired by existing formulations [32]: using CPLEX 12.7.1. The input data of the example include the fol-
minimize Z1 = Cmax (4) lowing: the processing times are generated following a U (1, 99) con-
tinuous uniform distribution as is usual in the flowshop scheduling
minimize Z2 = TCE = qijk x ijk + sqi (Cmax pijk x ijk ) literature; there is a set of three processing speeds factors
i j k i j k
= {1.2, 1, 0.8} ; the set of energy consumption factors is
(5) = {1.5, 1, 0.6} ; the power of each machine i in kw is i = 60, i ;
subject to the carbon dioxide conversion factor is 0.785 kgCO2 /kwh; and finally
the stand-by energy consumption factor of each machine i is
Cmax Cmj j (6) . Fig. 3 is a trade-off curve that shows the set of al-
i = 0.1, i
ternative solutions, or non-dominated solutions, of this illustrative ex-
C1j p1jk x1jk j
k (7) ample of GFSP for five sets of randomly generated data.
For the data set in Fig. 3, the alternative solutions reveal that a GFSP
Cij Ci 1, j + pijk xijk i {1};j is inherently a bi-objective scheduling problem which involves two
k (8) conflicting optimization criteria. To further understand how alternative
solutions differ from each other, we also depict in Fig. 3 the individual
Cij Cil + pijk xijk M i ;j, l ,j l
jl
(9) completion times and carbon emissions for 10 jobs at five specific al-
k
ternative solutions. The outcomes of job completion times and job
Cil Cij + pijk xijk M (1 jl ) i ;j, l ,j l carbon emissions are shown in Fig. 3a and Fig. 3c, respectively. Despite
k (10) the conflict of Cmax and TCE, one can argue that a GFSP has the fol-
lowing special property.
jl + lj =1 j, l ,j l (11)
Property 1. Selection of a slower job processing speed (or equivalently a
jl + lo + oj 2 j, l, o ,j l o (12) lower carbon emissions) does not necessarily result in an increase of Cmax of
the GFSP.
xijk = 1 i ;j
k (13) Property 1 can be shown by an illustrative example of two jobs j and
l that are processed on two consecutive machines i and i + 1 at the same
Cij 0, x ijk , jl {0, 1} i ;j, l ;k (14)
time. Assume that pij = 10 and pi + 1, l = 11 and = {1.2, 1, 0.8} . Fur-
The objective functions, Eqs. (4) and (5), minimize makespan and thermore, assume that job j starts its process on machine i at the instant
total carbon emissions over the planning horizon as the first objective of time 30 with an unfixed processing speed, whereas job l starts its
Z1 and the second objective Z2 of the problem, respectively. In Eq. (5), process on machine i + 1 at the instant of time 35 with a fixed pro-
total carbon emitted, TCE, is in kgCO2 . It is noticeable that the first term cessing speed 1.2. Then, one can argue that, no matter how quickly we
of the equation calculates the carbon emission of machines during their process job j on machine i, machine i + 1 cannot be empty before the
operational period and the second term of the equation calculates the instant of time 44.16. This means that considering the maximum
carbon emissions of machines during their idle period. floating for processing job j on machine i does not cause an increase in
With regard to the set of constraints, Inequality (6) guarantees the Cmax of the GFSP.

305
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Fig. 3. The set of alternative solutions of this base case example for five sets of randomly generated data.

The example in Fig. 3 only illustrates a GFSPs with no carbon re- the objective function, or equivalently we can consider Cmax as the
duction policy, and consequently we now examine the effect of major single criterion of the model. Furthermore, for the single carbon base-
carbon reduction policies on GFSPs. line, there is a unique threshold with respect to the amount of carbon
emitted throughout the manufacturing period. Equivalently, we denote
3.4. Modelling GFSPs with ToE, BoE and ET carbon reduction policies the value of this strict threshold by Q in kgCO2 as the maximum possible
amount of carbon emitted in each manufacturing period. This means a
The ToE carbon reduction policy is very common in practice. For solution can be feasible for the problem only if it satisfies the following
example, total carbon taxes are often applied for petrochemical-related additional constraints:
industries [35]. Note that an advantage of the ToE policy is that the
revenue raised from it can be used to fund investment in alternatives. qijk xijk + sqi (Cmax pijk x ijk ) Q
i j k i j k (17)
For example, the government can grant subsidies for more en-
vironmentally friendly public transport. For this policy, the goal is to By recalling Constraints (6)–(14), the MILP model of GFSPs with a
minimize the total carbon taxes of the flowshop over the entire plan- BoE carbon reduction policy is min {Cmax | s . t . (6)–(14) and (17)} . We
ning horizon. Obviously, the planning horizon for this case is the ma- label this model as P2-BoE due to the similarity between it and the
kespan since it determines the time taken for completion of all n jobs on model P2. In P2-BoE, Constraint (17) controls the total amount of
m machines. Therefore, the revised objective function of P1 is carbon emitted in each time interval (0, Cmax ) to guarantee it never
minimize Z1 = Cmax exceeds the threshold.
minimize Z2 = TCE (15) For comparison purposes, recall that the ET carbon reduction policy
is a combination of the first two policies. On the one hand, there is no
where is the amount of tax paid on each kilogram of carbon emission. carbon constraint in the ToE policy, and it allows manufacturers to
Note that, as an alternative to this objective function, we proposes the produce carbon dioxide in an unlimited manner while they are willing
following proportion-based objective function to strengthen the con- to pay its related penalties (respectively, taxes). Note these penalties are
vergence to alternative solutions as well as maintain a diversity of them only considered as a part of the objective function of the model P1-ToE.
by replacing with 1 . On the other hand, the BoE policy puts a strict carbon baseline on
emissions. In this case, strict carbon baselines are only considered as
minimize Cmax + (1 ) TCE (16)
constraints of the model P2-BoE. Clearly, each one of these policies has
In the aforementioned objective function, which is a weighted sum its own advantages and disadvantages since they can be referred to as
function or scalarization in the context of multi-objective optimization, two polar cases. Therefore, a hybrid ET policy is intended to capture
is a parameter between 0 and 1 that combines Cmax and TCE into a advantages of both. The reason behind this intuition is that the ET
single convex combination of objectives since 0 1. The objective policy has the carbon penalty and reward in the objective function and
function exclusively represents Cmax if = 1, and exclusively represents the carbon baseline in the constraints. Hence, this characteristic adds
TCE if = 0 . Taking Eq. (16) into account, the MILP model of GFSPs flexibility in comparison with other policies.
with a ToE carbon reduction policy is The ET policy leads to a situation in which a GFSP is subject to
min {Cmax and TCE| s . t . (6)–(14)} . We name this model P1-ToE for the lenient carbon baselines. Under this condition, the flowshop can be
sake of brevity. penalized (rewarded) for emitting more (less) than its baseline since the
A BoE carbon reduction policy is an alternative to a ToE carbon baseline is not strict. The trading of carbon is an international trade
reduction policy. Under this policy, instead of putting taxes on emis- with two goals, reduction of the air pollution caused by carbon emis-
sions, government places strict carbon baselines on emissions. The ad- sions and increasing profit of companies by selling their carbon credits
vantages of this policy is that it guarantees the carbon emission of a simultaneously. From a managerial point of view, the distinguishing
flowshop is certainly less than a given threshold, whereas the taxes on characteristic of the ET policy is that it is an encouragement scheme for
emissions policy is not able to offer such a guarantee on the amount of carbon reduction. Regardless of the total amount of carbon emitted in
carbon emitted in the flowshop, which may result in a larger and un- flowshop lines of companies, they can benefit from the ET policy - ei-
controlled carbon footprint. To analyze the BoE policy for GFSPs, it is ther by selling off surplus emission allowances to other companies, or
necessary to assume additional structure on the scheduling problem by buying additional emission allowances from other companies, or
under this policy. even trading off within operating plants of a single multinational
Unlike the ToE policy where a carbon tax function represents the company with operations in different countries - if it can increase their
carbon emissions as a component of the objective function, the carbon net profit (e.g., by decreasing Cmax ).
emissions should not exceed a given threshold for BoE policy. Let us consider the following objective function to shed light on a
Therefore, we can remove the environmentally related criterion from suitable model for the ET policy.

306
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

minimize Z1 = Cmax behaviours.

minimize Z2 = (TCE Q) (18) In practice, allocation of emissions allowances can employ different
mechanisms such as grandfathering and auctioning. For grandfathering
where is the amount of money paid (or received) per unit of carbon mechanism, firms receive free carbon allowances based on historic
credits. If , and Q are given values, then Eq. (18) is equivalent to Eq. emissions, which is represented as Q in the ET policy (or somewhat
(15) from a computational point of view. Further, it may be the case similar to Q in the BoE policy). In contrast, firms have to pay for carbon
that Eq. (18) is rewritten as the integration of the equation allowances in an auctioning mechanism, which is represented as in
Z2 = (e+ e ) and additional constraint TCE + e Q + e+, where the ET policy (somewhat similar to in the ToE policy). This ac-
auxiliary variables e , e+ 0 . One may look at this case and find it knowledges the link between carbon reduction policies and carbon al-
equivalent to Eq. (15) from a computational point of view. Due to this location mechanisms addressed in Zetterberg et al. [36].
equivalency, it is not necessary to compare extracted models compu-
tationally. Instead, we provide a policy-oriented environmental study of 3.5. Bi-objective optimization of GFSPs
GFSPs, which is intuitively acceptable for policy-makers who often do
not consider , and Q as given values. This appear a logical extension A solution to the bi-objective GFSP, say 1 defines both a feasible
of GFSPs since, for example, is affected by a regulating agency or permutation of n jobs, say 1, and an m-by-n feasible matrix of im-
market dynamics and it is often subject to volatility (e.g., see [1]). Note plemented processing speed factors for jobs on machines, say 1 (e.g.,
that it may be preferable to rewrite Eq. (18) as the following propor- an m-by-n matrix of the speed factor selected for each of n jobs to be
tion-based objective function (e.g., employing a linear scalarization processed by each of m machines). Generally speaking, we say solution
method) by replacing with 1 .
1 dominates another solution 2 if 1 is not worse than 2 under any of
minimize Cmax + (1 )(TCE Q) (19) the performance measures Z1 and Z2 , and 1 is strictly better than 2
under at least one of Z1 or Z2 . Non-dominated solutions are classified as
Here, the carbon emissions should not exceed a given threshold Q. This alternative solutions, which are detailed as follows:
characteristic resembles that of Eq. (18), with Constraint (17) of the
Definition 1. We say that 1 dominates another 2 and represent it by
model P2-BoE. However, the carbon baseline is not a mandatory
baseline in the ET policy. Instead, TCE Q represents the amount of 1 2 if Z1 ( 1) Z1 ( 2 ) and Z2 ( 1) Z2 ( 2 ) with one strict inequality. A
solution is named an alternative solution, if we cannot find any other
carbon credits sold or bought in the planning horizon Cmax if TCE Q is
where . Provided that is an alternative solution,
a negative or a positive value, respectively. Clearly, Constraint (17) is
Z = (Z1 ( ), Z2 ( )) represents its solution value.
not hard if the ET policy is applied for GFSPs (e.g., it can be replaced
with the soft constraint i j k
qijk Definition 1 is commonly used in the context of multi-objective
xijk + i sqi (Cmax pijk x ijk ) + e Q + e+ with auxiliary optimization, and has therefore been adopted for the purpose of carbon
j k
variables e and e+ that make it possible to violate the relaxed con- reduction in FSPs [29]. This definition can also be used to characterize
straint), whereas Constraint (17) is a hard constraint in P2-BoE that carbon reduction policies as follows:
cannot be violated. Interestingly, when Constraint (17) is not a binding Property 2. The set of alternative solutions is independent of financial
constraint for the case of ET policy, we have Q TCE units of unused parameters and although and affect solution values.
carbon credits. Therefore, we can sell them by the associate price as
the objective function of the ET policy shows. In addition, when Con- Proof. The fact that and only appear in objective functions of
straint (17) is a binding constraint, we are able to buy TCE Q units of models P1-ToE and P1P2-ET (respectively, Objective functions (15) and
carbon credits from the market. Although we buy TCE Q units of (18)) makes it clear that there is no alternative solutions set affected by
carbon credits with the price , this action may dramatically decrease these parameters. □
the objective function value. This is because units of carbon credits Property 3. The alternative solutions sets of models P1-ToE and P1P2-ET
are able to decrease Cmax in the objective function by relaxing Con- are not sensitive to and , whereas the optimal solution of P2-BoE is
straint (17). Therefore, we can conclude that the GFSP should be for- sensitive to Q.
mulated as the MILP model min {Cmax and (TCE Q)| s . t . (6)–(14)}
which we label as P1P2-ET due to its similarities with both MILP Proof. The proof can be presented in a structure similar to that of
models P1-ToE and P2-BoE. For the sake of brevity, Table 1 summarizes Property 2. We summarize it as follows:
these three MILP models according to the design features of the cor-
responding GFSPs under study. • Because is only part of Objective function (15), the alternative
Of course, the hybrid ET policy exhibits strong similarities to both solutions set of P1-ToE does not change by varying , although their
the ToE and BoE policies, especially from a modelling point of view of solution value changes.
shop scheduling problems. We are more likely to observe these math- • Because Q is the value of the right hand side of the Constraint (17),
ematical similarities when manufacturing companies use the ET policy, the optimal solution of P2-BoE changes by varying Q, although none
and this is due to the fact that , and Q are fixed values for given of solution values changes.
operational plans. However, we may rarely observe any similarities • Because is only part of Objective function (18), the alternative
between the ET policy and others for tactical plans. More explicitly, solutions set of P1P2-ET does not change by varying , although
, and Q are not given values of policies, and governmental policy- their solution value changes.
makers are in charge of determining their optimal values to affect • Because Q is only part of Objective function (18), the alternative

Table 1
Summary of design features of models P1-ToE, P2-BoE and P1P2-ET.

307
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

solutions set of P1P2-ET does not change by varying Q, although (low carbon-intensive industries), focusing on on-site emissions [40].
their solution value changes. □ Therefore, there is a possibility to classify these industry sectors based
on data in Table 2 (e.g., low carbon-intensive industries: = 12,
Properties 2 and 3 provides us with some intuitions about alter-
1 = 1.5, 2 = 1, 3 = 0.6 and = 12, 1 = 3, 2 = 1, 3 = 0.3 and
native solutions of the three models. We will use these properties in the
high carbon-intensive industries: = 75, 1 = 1.5, 2 = 1, 3 = 0.6
following section, when we implement models and provide numerical and = 75, 1 = 3, 2 = 1, 3 = 0.3.
results. Here, it should be noted that there are a fairly large number of studies
related to FSPs with different production data characteristics and mainly
4. Model implementations and numerical results focused on computational aspects of the problem. Therefore, our main
focus is neither on exploring the impact of production data nor on
4.1. Base case real-world data benchmarks computational procedures and algorithms. Instead, we concentrate on
environmental and financial data since they have a significant influence
Data collection is performed in three stages. As detailed in Table 2, on carbon emissions of GFSPs. In addition, we provide the sensitivity of
realistic production data and their sizes are easily collected from widely an optimal sequence to the input environmental and financial parameters
used benchmarks for FSPs since they are not related to sustainability in order to give clues about the structural nature of carbon reduction
issues. However, there are no recognized benchmark instances that policies. The reason why environmental and financial data attract our
exist for GFSPs. We generate environmental parameters in accordance main focus here is that they both play an important role in manu-
with other studies in the area of GFSP since they are still related to facturing as well as in service, as detailed in Zhang et al. [28] for GFSPs.
flowshop process models. We assume two levels of machine powers that Moreover, variations of the FSP in practice have to be supported by some
are 12 kw (e.g., continuous fusing heat press machines) and 75 kw (e.g., form of parametric or sensitivity analysis [13].
plastic injection molding machines). We test the models with two sets of
energy consumption factors 1 = {1.5, 1, 0.6} representing a set of fac- 4.2. Sensitivity analysis of carbon reduction models and numerical results
tors with low diversity, and 2 = {3, 1, 0.3} representing the case where
energy consumption factors are more diverse. Finally, financial para- Sensitivity analysis is popular in the domain of energy-efficient shop
meters are collected from general sustainability studies since they are scheduling from both energy-effective and cost-effective viewpoints.
more broadly relevant beyond GFSPs. For instance, the interval of For instance, Gong et al. [41] provided a sensitivity analysis for a shop
carbon price we select, (0, 0.050), is in line with the environmental scheduling problem related to a Belgium plastic bottle manufacturer, in
legislation in Australia where carbon is taxed at A$23 per tonne (0.023 which night and weekend shifts are lower-priced (higher-priced) per-
per kilogram) in 2012 rising by 5% each year until reaching a global iods for the electricity consumption (the labor) cost. Likewise, Zhang
price of A$36 per tonne (0.036 per kilogram) in 2022 [37]. In addition, et al. [42] targeted the optimal design of a shop scheduling problem
i j
qij3 and i j
qij1 represents the lower bound and the from cost-effectiveness and productivity enhancement. They performed
upper bound of carbon emissions, respectively. It is obvious that there is a sensitivity analysis on the problem to find the relationship between
no feasible schedule when Q < i j
qij3 since all jobs are pro- shop productivity and the main factors. Generally, sensitivity analysis is
cessed at their lowest speed, or equivalently lowest carbon emissions the task of exploring the behaviour of the final solution of an optimi-
level. Likewise, the baseline is meaningless when Q > i j
qij1 zation problem due to changes in the input parameters. In the previous
because all jobs can be processed at their highest speed, or equivalently section we provided a set of models to study the impact of different
highest carbon emissions level. carbon reduction policies on FSPs, assuming that input parameters were
In reality, carbon footprints vary across industries, and even be- given. This is not necessarily the case as solutions are possibly sensitive
tween different firms producing the same products, because the design to the changes in input parameters, meaning that even minor changes
and age of machines and their operating and maintenance practices can in them may have a significant impact on the final solution. This is
yield different carbon footprints. Specifically, the manufacturing sector compounded by the fact that, on one side, companies often have in-
is generally divided into 15 sub-sectors: chemicals, petroleum refining, sufficient information to assign suitable values to input parameters such
forest products, food and beverage, iron and steel, transportation as , and Q. On the other side, it is important for policy-makers to
equipment, plastics, fabricated metals, cement, alumina and aluminum, find the most efficient values of design parameters , and Q.
computers and electrical equipment, textiles, machinery, glass, and Therefore, in this study we focus on how to overcome this barrier.
foundries worldwide. In term of carbon emissions, the first half of the Here, we implement our models in OPL and employ CPLEX 12.7.1 to
sub-sectors have the highest emissions (high carbon-intensive in- solve P1-ToE, P2-BoE and P1P2-ET for GFSPs of a reasonable size.
dustries) and the second half of sub-sectors have the lowest emissions Considering data given in Table 2, we focus our analysis on the

Table 2
Summary of benchmark test problem parameters.
Parameter type Parameter name Parameter levels Based on

Production The size of processing time matrix 10 × 10 Vallada et al. [38]

Processing times in minutes U (1, 99) Vallada et al. [38]
Processing speed factor 1.2, 1, 0.8 Mansouri et al. [30]

Environmental Machine power in kilowatt 12, 75 Mansouri et al. [30]

Energy consumption factor 1 = {1.5, 1, 0.6} , 2 = {3, 1, 0.3} Mansouri et al. [30]
Stand-by energy factor 0.05 Ding et al. [29]
Conversion factor to CO2 0.785 Liu et al. [8]

Financial Carbon tax price per kilogram (0, 0.050) Fahimnia et al. [37]
Carbon baseline in kgCO2 ( i j qij3, i j qij1) Fang et al. [32]
Carbon credit price per kilogram (0, 0.050) Zakeri et al. [39]

308
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

numerical results for randomly generated instances. Tables 3–5, re- Further, results suggest that a carbon tax around A$0.030 gives the
spectively, illustrate numerical results of P1-ToE, P2-BoE and P1P2-ET most balanced ToE policy in Table 3. The reason behind this intuition is
for a particular case (e.g., i = 75 kW and i = { 1, 2 , 3} = {3, 1, 0.3} ). that, although TCE decreases sharply with an increase of Cmax at the
Let us now start by analyzing P1-ToE in Table 3 where the first beginning of the ToE policy, the overall upward trend loses its sharp-
column of the table provides the test instance ID with changing input ness gradually. Therefore, A$0.030 can be identified as the most ap-
variables for the purpose of sensitivity analysis, and the second column propriate since we cannot observe a high reduction in TCE with a
represents the value of the carbon tax . The baseline model, in which a higher although it considerably increases Cmax . Note that policy-ma-
company ignores the effect of the ToE policy on its flowshop models, can kers can use their subjective judgement to decide what carbon tax price
be specified by a triple in the third (Cmax ), fourth (TCE), and fifth ( TCE ) is an ideal choice, and this depends on their perception of the relative
columns. In contrast, when the company focuses its attention on the ef- importance of objectives Cmax and TCE.
fect of the ToE policy on its flowshop models, an alternative solutions set Table 3 is restricted to the case i = 75 kw and 2 = {3, 1, 0.3} as a
can be specified by a triple in the sixth, seventh and eighth columns, of representative of a GFSP where machines have a high energy con-
an efficient choice of the front for the model P1-ToE with both economic sumption as well as a high variety of energy consumption factors.
and environmental concerns. Finally, the gap between triples for each Consequently, the table is not able to capture the effect of carbon taxes
one of Cmax , TCE and TCE is reported in columns 9, 10 and 11 to on industries with different environmental parameters (e.g. with two
monitor the efficiency of solutions for each level of carbon tax. different types of machine powers 1 = 12 and 2 = 75 and energy
Analysis of P1-ToE in Table 3 from an operational point of view: it is consumption factors 1 = {1.5, 1, 0.6} and 2 = {3, 1, 0.3} ). Motivated
worth mentioning that companies often do not seek to decrease TCE by this reason and as a framework for different types of companies, the
unless, of course, it has a significant advantage, or at least no detri- trade-off curve in Fig. 4 informs how improving TCE and TCE relates
mental impact, on their financial performance. Therefore, their primary to deteriorating Cmax while moving along the curve for each combina-
focus will be Cmax (columns 3 and 6) and TCE (column 8 if a ToE is tion of i, i and . For the sake of simplicity, we used and instead
mandated), not TCE (columns 4 and 7) per se. As column 3 shows, the of i and i in this figure if all machines have identical powers and
problem can be converted to a single-objective optimization problem if energy consumption factors. Recall that the reason why the y-axis re-
Cmax is the only objective of the GFSP and this results in the maximum presents both TCE and TCE is that they are of interest to companies
TCE as shown in column 4. If this is the case, i.e. no ToE policy is and policy-makers, respectively. In the figure, the distinction between
implemented by policy-makers, there is no disincentive to produce TCE and its resulting cost lies in the fact that while a single curve be-
carbon emissions. But if a ToE policy is initiated, column 5 reveals that longs to TCE for different values of carbon tax, TCE s, on the other
minimization of Cmax may not be reasonable due to the negative in- hand, are not exclusive because they can be shown with different curves
clination of the trend of TCE . In this case, a trade-off arises between when is changing. In Fig. 4, one of the following four cases occurs to
Cmax and TCE , as shown in columns 6 and 7 in order to control the capture the operations of most likely industry sectors:
upward trend of TCE . As can clearly be seen, in order to keep the
overall trend of TCE stable, we must decrease TCE with negative (i) = 12, 1 = 1.5, 2 = 1, 3 = 0.6: it represents the overall struc-
consequences on Cmax . This means that we can cope with an increase in ture of low-carbon industries where the carbon emissions are low
the carbon tax if the objectives of the GFSP are well balanced. However, and within a narrow range between 938.62 kgCO2 and
this is not always the case, as it sometimes becomes possible to decrease 1246 kgCO2 , and therefore carbon reduction is not a major issue in
TCE without sacrificing Cmax . This is due to Property 1 of GFSPs that is this case. In other words, neither companies nor policy-makers are
mentioned in Section 3.3. As an example of this situation, simulta- motivated to invest in implementation of strict carbon reduction
neously consider tests 0 and 1 where Cmax remains 896.54 min although policies. This is because, by increasing , we never observe a sharp
TCE decreases to 9476.71 kgCO2. This is because of the possibility of decrease beyond the slope of curved dashed line showing balanced
keeping Cmax fixed by considering other permutations of jobs with more choices.
efficient choice of job speed factors as well. This means a A$0.0067 cost (ii) = 12, 1 = 3, 2 = 1, 3 = 0.3: similar to the previous case, this
saving per kilogram with the same competitiveness of the shop. In case is related to low-carbon industries. However, here an in-
columns 9, 10 and 11 of Table 3, we have shown how a balanced choice effective schedule can increase the chance of a high carbon emis-
of job permutations and their speed factors can provide an effective sion happening due to the wide range of carbon emissions (from
production plan of the GFSP. These columns of the table support the 651.22 kgCO2 to 2258 kgCO2 ). Hence, the policy-maker’s decision
intuition that saving carbon costs necessarily entails loss of some equity is naturally to increase carbon taxes in comparison with the pre-
of Cmax . vious case. Given this incentive, companies can schedule their
Analysis of P1-ToE in Table 3 from a tactical point of view: as discussed flowshops to save more carbon emissions in this scenario, as we
earlier, it is important to know what is the best decision to be made observe a sharper decrease beyond the slope of the curved dashed
under a given from a company perspective. So, we should always limit line of balanced choices.
analysis to one row of Table 3 for an operational analysis since is a (iii) = 75, 1 = 1.5, 2 = 1, 3 = 0.6: this case reflects high-carbon
given parameter for the previous case. However, in contrast, here industries where all machines potentially cause high carbon
plays the role of a decision variable for policy-makers because they are emissions (the range of carbon emissions is between
often interested to determine a carbon tax value that keeps different 4144.66 kgCO2 and 6558 kgCO2 ). This means, on the one hand, a
companies economically effective, while decreasing the level of their high tax cost for the company that is unavoidable and, on the other
carbon emissions. This therefore means that all rows of Table 3 should hand, a high carbon emissions that cannot be significantly con-
be analyzed simultaneously, to find the optimal value of from a trolled by policy-makers. Thus, we observe a slightly sharp de-
policy-maker’s point of view. Examining Table 3, this time vertically, crease beyond the slope of the curved dashed line of balanced
we can see that considering A$0.002 instead of A$0 results in no in- choices in this case.
crease in Cmax . In other words, a first glance suggests that policy-makers (iv) = 75, 1 = 3, 2 = 1, 3 = 0.3: similar to Case (iii), this case
should assign a tax to the carbon emission even if it targets a small reflects high-carbon industries. However, since we observe a wide
share of a company’s budget. This situation is a nice example of how the range of carbon emissions ranging between 2247.42 kgCO2 and
solution is actually a more energy efficient one (with reduced TCE) with 12882 kgCO2 , we are able to decrease carbon emissions sig-
the advantage for the company that Cmax is no different. In more detail, nificantly if machines are well managed. Note that here, an in-
recalling Property 1, it is effectively encouraging solutions with the effective schedule may result in the peak carbon emission amongst
same Cmax to be explored and selecting the one with the reduced TCE. the four cases, whereas an effective schedule may result in a

309
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Table 3
Result of solving uniformly generated test instances of GFSPs with ToE policy.
Test Carbon tax No ToE policy consideration ToE policy consideration Efficiency gap of ToE policy

# ($/kgCO2) Cmax (min) TCE (kg) TCE ($) Cmax (min) TCE (kg) TCE ($) Cmax (%) TCE (%) TCE (%)

0 0.000 896.54 12882 0.00 896.54 12882 0.00 0.00 0.00 0.00
1 0.002 896.54 12882 25.66 896.54 9476.71 18.95 0.00 26.43 100
2 0.004 896.54 12882 51.33 896.65 9156.75 36.62 0.01 28.92 93.25
3 0.006 896.54 12882 76.99 897.26 9055.72 54.33 0.08 29.70 48.36
4 0.008 896.54 12882 102.66 917.54 8064.56 64.51 2.34 37.40 18.74
5 0.010 896.54 12882 128.32 924.68 7735.02 77.35 3.14 39.95 19.90
6 0.012 896.54 12882 153.98 939.05 7191.92 86.30 4.74 44.17 11.57
7 0.014 896.54 12882 179.65 949.76 6842.10 95.78 5.94 46.89 10.98
8 0.016 896.54 12882 205.31 974.62 6137.62 98.20 8.71 52.36 4.53
9 0.018 896.54 12882 230.98 985.04 5870.36 105.66 9.87 54.43 6.60
10 0.020 896.54 12882 256.64 996.78 5598.35 110.97 11.18 56.54 3.83
11 0.022 896.54 12882 282.30 1009.84 5201.15 115.22 12.64 59.62 3.06
12 0.024 896.54 12882 307.97 1026.00 4947.60 118.74 14.44 61.59 4.17
13 0.026 896.54 12882 333.63 1033.30 4802.90 124.88 15.25 62.72 3.53
14 0.028 896.54 12882 359.30 1043.80 4597.00 128.72 16.43 64.31 4.06
15 0.030 896.54 12882 384.96 1052.55 4455.04 133.95 17.40 65.42 2.86
16 0.032 896.54 12882 410.62 1060.70 4305.82 137.78 18.31 66.57 2.75
17 0.034 896.54 12882 436.29 1074.12 4090.83 139.09 19.81 68.24 2.65
18 0.036 896.54 12882 461.95 1081.22 3988.87 143.06 20.60 69.04 1.91
19 0.038 896.54 12882 487.62 1091.01 3877.16 145.76 21.69 69.90 1.89
20 0.040 896.54 12882 513.28 1102.09 3713.51 148.54 22.93 71.17 1.72
21 0.042 896.54 12882 538.94 1110.15 3613.91 151.79 23.83 71.95 2.09
22 0.044 896.54 12882 564.61 1121.18 3480.64 152.65 25.06 72.98 1.57
23 0.046 896.54 12882 590.27 1133.50 3335.61 153.44 26.43 74.11 1.52
24 0.048 896.54 12882 615.94 1142.55 3295.53 156.01 27.44 74.42 1.67
25 0.050 896.54 12882 641.60 1152.50 3166.66 158.33 28.55 75.42 3.49

Fig. 4. The ToE carbon reduction policy from operational and tactical points of view: four different scenarios.

carbon emission less than both Cases (ii) and (iii). This suggests a additional columns in Table 4 requiring explanation. The second and
need for a high carbon tax rate from policy-makers, forcing com- third columns represent the range of the carbon baseline Q. As stated in
panies to reduce their carbon emissions as the dramatically sharp Section 4.1, Q should be selected from the interval
decrease beyond the slope of the curved dashed line indicates. ( i j
qij3, i j
qij1) . For this purpose, the second and third
columns show the percentage of the baseline tightness defined by
Let us now analyze P2-BoE in Table 4 where most columns are si- i j qij1 Q
in an increasing order.
milar to Table 3 regardless of carbon taxes. However, there exist two i j qij1 i j qij3

310
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Analysis of P2-BoE in Table 4 from an operational point of view: under strict alternative solutions. Accordingly, each time that we run the
this condition, the problem is converted into a single-objective GFSP, model by CPLEX, we select an appropriate solution of the GFSP for the
which only involves Cmax . In other words, TCE only appears as a hard case i = 75 kw and i = { 1, 2 , 3} = {3, 1, 0.3} . Following the same
Constraint (17), which is often a binding constraint as TCE Q de- approach as presented in GFSPs with ToE and BoE policies, we should
duced from columns 3 and 5 of Table 4 illustrates. So, simply, we analyze results of Table 5 for P1P2-ET. The results obtained in Table 5
should minimize Cmax given a threshold for TCE. At first glance, this show that a ET policy in general is more suitable than ToE and BoE
policy may be interesting since it decreases the complexity of the pro- policies, not only with respect to the flexibility of solutions, but also
blem and it puts no financial pressure on the system. However, in benefits to manufacturers and policy-makers. In short, by the analysis of
contrast with the ToE policy, the disadvantage of the BoE policy from a Tables 3–5, one can see that ET policy is more effective than ToE policy,
company point of view is that it has no flexibility in the event of the while it respects the limits enforced by the BoE policy. Further, there is
need for extra carbon emissions. evidence in Table 5 of the encouraging behaviour of the ET policy, as
Analysis of P2-BoE in Table 4 from a tactical point of view: For policy- TCE Q often takes a large negative value to increase the revenue of
makers, the challenge is how to determine the decision variable Q, in companies and it has a greater impact in reduction of carbon footprints.
contrast with the previous case where Q is a given value. Considering Even when the value of increases, we witnessed the increase of rev-
26 tests in Table 4 simultaneously, we observe that the baseline is not enue in most cases. This may be interpreted as a sign that ET policy
tight enough to decrease TCE in an effective manner when it is con- could be more welcomed by companies. In order to provide a detailed
sidered 28% tight or less, and this means that Property 1 of GFSPs is not analysis of P1P2-ET in Table 5, we refer to Fig. 6 for a visualization of
completely implemented under this condition. Consequently, in this GFSPs with ET policy.
sense, the BoE policy is worse than the ToE policy where only a small Analysis of P1P2-ET in Table 5 from an operational point of view: A
value of tax, e.g., A$0.002, encourages companies to decrease their saddle-shaped curve is formed by the intersection of and Q in Fig. 6a.
carbon emissions considerably. However, the BoE policy is potentially Recall that and Q are input parameters of GFSP from a company’s
straightforward to be implemented for GFSPs in practice. In more de- perspective although they play the role of decision values from a policy-
tail, we can simply determine the optimal Q by looking at columns 8 maker’s perspective. For the sake of generalization, the saddle-shaped
and 9. What can be observed is that Cmax experiences a quadratic curve reflects the situation when both and Q have the same priority
growth in column 8, whereas TCE linearly decreases in column 9. for a GFSP (e.g., = 0.5). As we see from the nature of the ET policy,
Therefore, a baseline with 80% tightness reflects almost the most ap- the harmony between and Q is the reason why we obtain a saddle-
propriate Q, 4374.34 kgCO2, to induce the desired sustainability level. shaped set of solutions here. In Fig. 6a, once the coordinates of two
As a roadmap for policy-makers, the set of trade-off curves in Fig. 5 parameters and Q are known, the z-axis shows the state of the optimal
assists them in determining the optimal baseline for each specific sce- solution of the GFSP. Following that, we can map this state to the set of
nario, considering both economic limitations and carbon footprint alternative solutions to find the decision related to companies. Notice
standards. In this figure, we show the suggested baselines with dotted the saddle point in this figure is obtained at the breakpoint in which the
circles, which are used to compare four different scenarios. We should slope of curve turns from increasing to non-increasing. This point is
here emphasize that policy-makers can use their subjective judgement important so that a positive slope corresponds to a situation in which a
to decide which baseline is an ideal choice. As we can see, the effect of a company should buy carbon credits, while a negative slope corresponds
higher set of machine powers in GFSPS, 75 kw, is that it shifts up the to a situation where a company should sell carbon credits as a revenue
trade-off curve with no considerable changes in range of the carbon opportunity.
emissions if 1 = 1.5, 2 = 1 and 3 = 0.6. Consequently, we are not To make our analysis more manageable, we should say that Fig. 6a
able to significantly decrease carbon emissions without causing eco- maps out the details for the optimal solution through Figs. 6a–f. Figs. 6a
nomic damage to companies by significantly increasing Cmax . It is and 5b together make it clear why Fig. 6a has a saddle shape instead of
possible that this does not work properly due to the narrow range of a flat surface. Moreover, in comparison to Fig. 6d, we expect to observe
accepted carbon emissions. Subsequently, our suggestion for policy- twisting of (TEC Q ) in -axis of Fig. 6c due to its similarity to Fig. 4.
makers is not to assign tight carbon emission baselines to companies Finally, Fig. 6d can be considered as the intersection of Figs. 6e and 6f,
with such process scenarios. However, our analysis strongly suggests which determines whether or not a company should buy credits. In
that policy-makers orient their policy in the opposite direction when other words, TEC Q tends to have a linearly negative slope in Q-axis
the carbon emissions from a company’s operations span a wide variety of Fig. 6d if TEC takes a constant value in Fig. 6e and Q has a linearly
depending on their choices. This case is graphically represented by the increasing slope in Q-axis of Fig. 6f.
green-coloured curve in Fig. 5. When comparing this curve with the Analysis of P1-ET in Table 5 from a tactical point of view: The saddle-
purple-coloured curve, we can clearly see the importance of an tight shaped curve of Fig. 6 provides policy-makers an opportunity to use
carbon emissions baseline. More precisely, considering the same Cmax , a and Q as two integrated options for returning to a balanced policy. With
tight Q results in lower carbon emissions whereas a loose Q results in this view, the following features are evident from the curve: the Q-axis
higher carbon emissions in comparison with the purple-coloured curve shows that the baseline has no effect on the performance of GFSPs if we
of GFSPs. apply no ET policy, = 0 , but > 0 is sufficient to see the effect of Q on
The analysis so far has shed light on the strengths and weaknesses of the solution values of the GFSP. We observe that loose baselines result
each of the ToE and BoE policies. We attempted to interpret outcomes in smaller solution values, whereas tight baselines gives us larger so-
of different scenarios cautiously, in a formative rather than judgemental lution values by gradually increasing carbon credits price in the market.
way. Moreover, the obtained results of GFSPs with ToE and BoE policies This effect can be seen in the peak of , A$0.050 per kilogram of CO2
are interpreted in the context of green shop scheduling and the context where the maximum solution value 577.82 occurs when Q is 100% tight
of the energy policy making in order to be attractive to both operations and it reduces to 322.59 when Q is 0% tight. Policy-makers are able to
research and energy strategy communities. However, it seems reason- directly control Q, whereas = 0 is a market-based price that they can
able to combine ToE and BoE policies into an integrated policy, the ET indirectly control. Thus, the ideal value of Q is when it is 50% tight,
policy, in order to increase (or decrease) their beneficial (or damaging) Q = 7564.71 kgCO2 if we have to select a global Q applied for all GFSPs.
impacts on GFSPs. Of course if we consider a Q tighter than 50% it decreases the total TCE,
Let us analyze P1P2-ET in Table 5 where most columns are similar but it may have economic impacts on companies and this does not keep
to Tables 3 and 4. We limit the test to five carbon credit prices and a Q < 7564.71 kgCO2 as the priority of policy-makers without providing
five carbon emissions baselines Q to make the table as compact and enough infrastructures. Returning to the curve, if there is no force to
informative as possible. Each solution in the table belongs to the set of select a global Q, we can see that a very loose Q still encourages the

311
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Table 4
Result of solving uniformly generated test instances of GFSPs with BoE policy.
Test Carbon baseline Carbon baseline No BoE policy consideration BoE policy consideration Efficiency gap of BoE policy

# (%) (kgCO2) Cmax (min) TCE (kg) Cmax (min) TCE (kg) Cmax (%) TCE (%)

0 0 12882 896.54 12882 896.54 12712.16 0.00 0.00

1 4 12456.62 896.54 12882 896.54 12433.55 0.00 3.11
2 8 12031.23 896.54 12882 896.54 12017.00 0.00 6.35
3 12 11605.85 896.54 12882 896.54 11563.11 0.00 9.89
4 16 11180.47 896.54 12882 896.54 11128.85 0.00 13.27
5 20 10755.08 896.54 12882 896.54 10754.75 0.00 16.19
6 24 10329.70 896.54 12882 896.54 10321.41 0.00 19.57
7 28 9904.32 896.54 12882 896.54 9902.17 0.00 22.83
8 32 9478.93 896.54 12882 896.57 9475.87 0.01 26.15
9 36 9053.55 896.54 12882 897.47 9048.07 0.10 29.49
10 40 8628.17 896.54 12882 906.52 8628.03 1.11 32.76
11 44 8202.78 896.54 12882 915.05 8198.19 2.06 36.11
12 48 7777.40 896.54 12882 924.18 7776.56 3.08 39.40
13 52 7352.02 896.54 12882 935.60 7351.83 4.36 42.71
14 56 6926.64 896.54 12882 948.14 6926.29 5.76 46.02
15 60 6501.25 896.54 12882 962.46 6497.25 7.35 49.37
16 64 6075.87 896.54 12882 978.23 6074.89 9.11 52.66
17 68 5650.49 896.54 12882 995.19 5650.25 11.00 55.97
18 72 4815.39 896.54 12882 1014.24 5224.90 13.13 59.28
19 76 4799.72 896.54 12882 1033.58 4799.42 15.29 62.60
20 80 4374.34 896.54 12882 1057.54 4373.69 17.96 65.92
21 84 3948.95 896.54 12882 1085.36 3948.45 21.06 69.23
22 88 3523.57 896.54 12882 1119.17 3523.20 24.83 72.54
23 92 3098.19 896.54 12882 1163.08 3097.25 29.73 75.86
24 96 2672.80 896.54 12882 1231.52 2672.10 37.36 79.18
25 100 2247.42 896.54 12882 1344.81 2247.41 50.00 82.49

Fig. 5. The BoE carbon reduction policy from operational and tactical points of view: four different scenarios.

GFSP to decrease its carbon emissions as much as possible in that it is a By similar arguments, we can see that the behaviour of ET policy
good source of revenue (e.g., a 0% tight Q with the highest , A$0.050 under four scenarios is close to Fig. 5. However, in contrast to the ToE
per kilogram of CO2, gives the maximum revenue to a company by policy, the horizontal axis here can take negative values of (TCE Q ) by
decreasing TCE as much as possible). In contrast, a very tight Q forces using tight Qs. Thus, for the sake of brevity, here we analyze only the
the GFSP to decrease the carbon emissions as much as possible in that it issues associated with different values of Q. The ultimate goal of this
decreases the peak carbon cost (e.g., a 100% tight Q with the highest analysis is to give policy-makers the knowledge for appropriate decision
forces the GFSP to decrease TCE either by switching the job processing making when carbon credits exist in the market of different industry
speeds or by buying carbon credits). sectors. Note Fig. 6a has interesting implications in practice, since it allows

312
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

us to obtain a negative solution value if we sell enough carbon credits. 4.3. Insights and discussions: the comparison of carbon reduction policies in
Further, the curve in Fig. 6a can be easily generalized, as depicted in Fig. 7 GFSPs
for the following five situations: (a) fully tight Q: highest priority for Cmax
(b) slightly tight Q: the main priority for Cmax (c) moderate Q: the same Here, four different versions of randomly generated test instances
priority for Cmax and (TCE Q ) (d) slightly loose Q: the main priority for were initially tested for GFSPs with three different carbon reduction
(TCE Q ) (e) fully tight Q: the highest priority for (TCE Q ). For the policies. The analysis of policies is based on real-world data on carbon
sake of clarity, we show a left side view of the figure for five situations (a emissions released by the Australian government. On the one hand, the
two-dimensional view of curves) instead of the comprehensive overall ET policy offers an encouraging policy, whereas the ToE policy is a
view of Fig. 6a (a three-dimensional view of the curve). discouraging policy. More particularly, the saddle-shaped curve of

Table 5
Result of solving uniformly generated test instances of GFSPs with ET policy.
Test (#) Carbon price ($/kgCO2) Carbon baseline (%) Cmax (min) TCE (kgCO2) TCE Q (kg) (TCE Q) ("$")

1 0.000 0 896.54 12882 0.00 0.00

2 0.000 25 896.54 12882 2658.64 0.00
3 0.000 50 896.54 12882 5317.30 0.00
4 0.000 75 896.54 12882 7975.93 0.00
5 0.000 100 896.54 12882 10634.58 0.00

6 0.0125 0 896.60 9157.21 −3725.91 −46.56

7 0.0125 25 896.60 9157.21 −1067.22 −13.33
8 0.0125 50 896.60 9157.21 1593.48 19.90
9 0.0125 75 896.60 9157.21 4251.50 53.13
10 0.0125 100 896.60 9157.21 6909.35 86.36

11 0.025 0 924.73 7735.55 −5147.06 −128.67

12 0.025 25 924.73 7735.55 −2488.33 −62.20
13 0.025 50 924.73 7735.55 170.84 4.25
14 0.025 75 924.73 7735.55 2829.20 70.72
15 0.025 100 924.73 7735.55 5488.11 137.19

16 0.0375 0 1133.62 3335.47 −9547.08 −358.02

17 0.0375 25 1133.62 3335.47 −6889.83 −258.32
18 0.0375 50 1133.62 3335.47 −4230.52 −158.63
19 0.0375 75 1133.62 3335.47 −1572.39 −58.93
20 0.0375 100 1133.62 3335.47 1087.88 40.76

21 0.050 0 1344.81 2247.88 −10634.26 −531.72

22 0.050 25 1344.81 2247.88 −7975.21 −398.79
23 0.050 50 1344.81 2247.88 −5317.03 −265.86
24 0.050 75 1344.81 2247.88 −2659.22 −132.93
25 0.050 100 1344.81 2247.88 0.00 0.00

Fig. 6. An example of ET carbon reduction policy ( = 0.5, = 75, 1 = 3, 2 = 1, 3 = 0.3).

313
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

Fig. 7. The ET carbon reduction policy under different priority settings.

Fig. 6 shows that a high satisfies some GFSPs, but all types of GFSPs model to decrease their carbon emissions. Instead, they can prevent
are likely to express their dissatisfaction with a ToE policy as presented cost-loaded schedules through minimizing TCE and (TEC Q ) in
in Fig. 4. On the other hand, the ET policy is not as baseline sensitive as GFSPs. Also, from the perspective of policy-makers, they can observe
the BoE policy as depicted in the saddle-shaped curve of Fig. 6. More that decision variables , Q and help them to initiate, execute and
precisely, the estimation of Q is hard because there is so much un- monitor Cmax and TCE of flowshop models in the course of each scenario
certainty in each industry sector (see e.g. Fig. 5). Further, fixing a value and its corresponding policy. The outcome of evaluation in most cases
for carbon emissions baselines seems unwise because they typically has shown that an Emissions Trading (ET) policy yields a stronger
involve many dynamic aspects. In order to prevent market abuse (e.g. market-based policy. However, adding two parameters and Q to the
by a very loose Q which makes it impossible to have an environmentally formulation not only strengthens the policy but also complicates its
friendly GFSP) and market misconduct (e.g. by a very tight Q which solutions due to the trade-off between parameters.
makes it impossible to have an effective GFSP), we can replace a BoE While beyond the scope of this initial study, our future work will
policy with a ET policy for the GFSP. This helps to identify threats and also consider several additional practical considerations. Firstly, it is
to mitigate potential damage in term of carbon emissions. However, we possible to face carbon tax functions with complex structures such as
still need to have a precise Q since it may unfairly cause additional nonlinearity, and this would necessitate revising P1-ToE, as stated in
economic costs for some companies. Ma et al. [43]. Secondly, baselines in P2-BoE that change from period to
Another feature that differentiates the ET policy from ToE and BoE period characterize many real-life GFSPs. For instance, Shrouf et al.
policies is the flexibility of this policy. Tables 3 and 4 show that ToE and [44] addressed a single-machine scheduling problem with varying en-
BoE policies for a GFSP can yield very different performances. On the ergy price based on the time of use in the Italian electricity market. For
one hand, the ToE policy acts as an infeasible schedule for BoE where this problem, the hourly electricity prices are available online the day
Q = 0 and the hard constraint (17) is always violated. Therefore, the before at www.Mercatoelettrico.org. An extension of their study can be
constraint (17) is removed by putting it in the objective function as a to identify the effect of periodic carbon baselines on other global
carbon-dependent tax for the violation of the relaxed constraint. On the electricity markets. In future work we can use the same framework to
other hand, the BoE policy acts in the opposite direction as ToE: the study a GFSP with varying energy price, as the hourly Australian gas
hard constraint (17) is never violated. To take into account both these and electricity prices are available online at www.aemo.com.au.
extreme policies, we can apply the ET policy to GFSPs, since ToE and Thirdly, it is worth noting that the model P1P2-ET, along with some
BoE polices act like special cases at the extreme points of Q in the ET practical assumptions, creates a mechanism for analyzing a variety of
policy. Table 5 shows that the treatment of the ET policy is similar to a more complex models. For instance, a variation of P1P2-ET is when
ToE policy when 0 and Q is very tight (100% tight), whereas the there are upper bounds on the number of carbon credits sold or bought.
treatment of the ET policy is similar to a BoE policy when 0 and Q Another variation of P1P2-ET occurs when the price of buying and
is very loose (0% tight). Therefore, the ET policy yields a trade-off selling of carbon credits are not same in the market. Finally, a sophis-
between and Q to solve GFSPs in a flexible and integrated manner. ticated case occurs when the planning horizon is composed of multiple
periods and each period has its own price of selling and buying carbon
5. Conclusions and future research credits, as stated in [1]. Studying the impact of carbon reduction po-
licies for these variations will assist policy-makers to provide the most
This paper has studied environmental issues in green flowshop effective green regulation so critical for many industry sectors in the
scheduling problems to capture both operational and tactical planning future.
decisions, mainly with regard to environmental and energy themed
contributions. A manufacturer should schedule its flowshop in a way to Acknowledgement
satisfy both demands and regulations, while a policy-maker should
justify parameters in a way that maximizes carbon reduction with the The authors would like to thank the anonymous referees for their
least harm to industry sectors who need to remain globally competitive. constructive suggestions and comments. The authors also acknowledge
Accordingly, we have applied such double justification to each set of support of the Australian Research Council Center of Excellence for
solutions generated in this study. Moreover, the emphasis has been Mathematical and Statistical Frontiers (ACEMS).
placed on Australia’s carbon reduction policies for numerical tests. We
have considered randomly generated test instances and derived the References
optimal solution sets for each of three carbon reduction policies. The
sensitivity analysis indicates that an optimal schedule can provide the [1] Benjaafar S, Li Y, Daskin M. Carbon footprint and the management of supply chains:
insights from simple models. IEEE Trans Autom Sci Eng 2013;10(1):99–116.
opportunity for manufacturers to save money or even generate addi- [2] Bayramusta M, Nasir VA. A fad or future of IT?: a comprehensive literature review
tional revenue from each policy. This means that they do not necessa- on the cloud computing research. Int J Inform Manage 2016;36(4):635–44.
rily need to adopt a cleaner production technology in their flowshop [3] Maroufmashat A, Sattari S, Roshandel R, Fowler M, Elkamel A. Multi-objective

314
M. Foumani and K. Smith-Miles Applied Energy 249 (2019) 300–315

optimization for design and operation of distributed energy systems through the [24] Manier M-A, Bloch C. A classification for hoist scheduling problems. Int J Flex
multi-energy hub network approach. Ind Eng Chem Res 2016;55(33):8950–66. Manuf Syst 2003;15(1):37–55.
[4] Suhariyanto TT, Wahab DA, Rahman MNA. Multi-Life Cycle Assessment for sus- [25] Kuntay I, Xu Q, Uygun K, Huang Y. Environmentally conscious hoist scheduling for
tainable products: a systematic review. J Clean Prod 2017;165(1):677–96. electroplating facilities. Chem Eng Commun 2006;193(1):273–92.
[5] Chen Q, Gu Y, Tang Z, Wei W, Sun Y. Assessment of low-carbon iron and steel [26] Wang S, Zhu Z, Fang K, Chu F, Chu C. Scheduling on a two-machine permutation
production with CO2 recycling and utilization technologies: a case study in China. flow shop under time-of-use electricity tariffs. Int J Prod Res 2018;56(9):3173–87.
Appl Energy 2018;220(1):192–207. [27] He S, Yin J, Zhang B, Wang Z. How to upgrade an enterprise’s low-carbon tech-
[6] Giret A, Trentesaux D, Prabhu V. Sustainability in manufacturing operations sche- nologies under a carbon tax: the trade-off between tax and upgrade fee. Appl Energy
duling: a state of the art review. J Manuf Syst 2015;37(1):126–40. 2018;227(1):564–73.
[7] Zhu Z-S, Liao H, Cao H-S, Wang L, Wei Y-M, Yan J. The differences of carbon in- [28] Zhang H, Zhao F, Fang K, Sutherland JW. Energy-conscious flow shop scheduling
tensity reduction rate across 89 countries in recent three decades. Appl Energy under time-of-use electricity tariffs. CIRP Ann – Manuf Technol 2014;63(1):37–40.
2014;113(1):808–15. [29] Ding J-Y, Song S, Wu C. Carbon-efficient scheduling of flow shops by multi-objec-
[8] Liu C, Yang J, Lian J, Li W, Evans S, Yin Y. Sustainable performance oriented op- tive optimization. Euro J Oper Res 2016;248(3):758–71.
erational decision-making of single machine systems with deterministic product [30] Mansouri SA, Aktas E, Besikci U. Green scheduling of a two-machine flowshop:
arrival time. J Clean Prod 2014;85(1):318–30. trade-off between makespan and energy consumption. Euro J Oper Res
[9] Fan JH, Todorova N. Dynamics of China’s carbon prices in the pilot trading phase. 2016;248(3):772–88.
Appl Energy 2017;208(1):1452–67. [31] Fang K, Uhan NA, Zhao F, Sutherland JW. A new approach to scheduling in man-
[10] Fang G, Tian L, Liu M, Fu M, Sun M. How to optimize the development of carbon ufacturing for power consumption and carbon footprint reduction. J Manuf Syst
trading in China-Enlightenment from evolution rules of the EU carbon price. Appl 2011;30(4):234–40.
Energy 2018;211(1):1039–49. [32] Fang K, Uhan NA, Zhao F, Sutherland JW. Flow shop scheduling with peak power
[11] Australian Department of Environment and Energy, 2017. < www.environment. consumption constraints. Ann Oper Res 2013;206(1):115–45.
gov.au/climate-change/government/emissions-reduction-fund/about/safeguard- [33] Gahm C, Denz F, Dirr M, Tuma A. Energy-efficient scheduling in manufacturing
mechanism > . companies: a review and research framework. Euro J Oper Res
[12] Ding J-Y, Song S, Gupta JND, Wang C, Zhang R, Wu C. New block properties for 2016;248(3):744–57.
flowshop scheduling with blocking and their application in an iterated greedy al- [34] Yenisey MM, Yagmahan B. Multi-objective permutation flow shop scheduling pro-
gorithm. Int J Prod Res 2016;54(16):4759–72. blem: literature review, classification and current trends. Omega
[13] Pinedo M. Scheduling, theory, algorithms, and systems. Springer International 2014;45(1):119–35.
Publishing; 2016. https://doi.org/10.1007/978-3-319-26580-3. [35] Lee CF, Lin SJ, Lewis C, Chang YF. Effects of carbon taxes on different industries by
[14] Minella G, Ruiz R, Ciavotta M. A review and evaluation of multiobjective algo- fuzzy goal programming: a case study of the petrochemical-related industries.
rithms for the flowshop scheduling problem. INFORMS J Comput Taiwan. Energy Policy 2007;35(8):4051–8.
2008;20(3):451–71. [36] Zetterberg L, Wrake M, Sterner T, Fischer C, Burtraw D. Short-Run allocation of
[15] Pannell DJ, Alston JM, Jeffrey J, Buckley YM, Vesk P, Rhodes JR, et al. Policy- emissions allowances and long-term goals for climate policy. Ambio
oriented environmental research: what is it worth? Environ Sci Policy 2012;41(1):23–32.
2018;86(1):64–71. [37] Fahimnia B, Sarkis J, Choudhary A, Eshragh A. Tactical supply chain planning
[16] Yu S, Wei Y-M, Guo H, Ding L. Carbon emission coefficient measurement of the under a carbon tax policy scheme: a case study. Int J Prod Econ
coal-to-power energy chain in China. Appl Energy 2014;114(1):290–300. 2015;164(1):206–15.
[17] Yan J, Zhao T, Kang J. Sensitivity analysis of technology and supply change for CO2 [38] Vallada E, Ruiz R, Framinan JM. New hard benchmark for flowshop scheduling
emission intensity of energy-intensive industries based on input–output model. problems minimising makespan. Euro J Oper Res 2015;240(3):666–77.
Appl Energy 2016;171(1):456–67. [39] Zakeri A, Dehghanian F, Fahimnia B, Sarkis J. Carbon pricing versus emission
[18] Mavromatidis G, Orehounig K, Carmeliet J. Uncertainty and global sensitivity strading: a supply chain planning perspective. Int J Prod Econ
analysis for the optimal design of distributed energy systems. Appl Energy 2015;164(1):197–205.
2018;214(1):219–38. [40] Brueske S, Sabouni R, Zach C, Andres H. U.S. manufacturing energy use and
[19] Shao C, Ding Y, Wang J. A low-carbon economic dispatch model incorporated with greenhouse gas emissions analysis; 2012. < https://www.energy.gov/sites/prod/
consumption-side emission penalty scheme. Appl Energy 2019;238(1):1084–92. files/2013/11/f4/energy_use_and_loss_and_emissions.pdf > .
[20] Shabtay D, Steiner G. A survey of scheduling with controllable processing times. [41] Gong X, Van der Wee M, Pessemier TD, Verbrugge S, Colle D, Martens L, et al.
Disc Appl Math 2007;155(13):1643–66. Integrating labor awareness to energy-efficient production scheduling under real-
[21] Uruk Z, Gultekin H, Akturk MS. Two-machine flowshop scheduling with flexible time electricity pricing: an empirical study. J Clean Prod 2017;168(1):239–53.
operations and controllable processing times. Comput Oper Res [42] Zhang J, Fang X, Qi L. Sensitivity analysis based method in single-robot cells cost-
2013;40(2):639–53. effective design and optimization. Robot Comput-Integr Manuf 2016;38(1):9–15.
[22] Mor B, Mosheiov G. Batch scheduling of identical jobs with controllable processing [43] Ma X, Ji P, Ho W, Yang C-H. Optimal procurement decision with a carbon tax for
times. Comput Oper Res 2014;41(1):115–24. the manufacturing industry. Comput Oper Res 2018;89(1):360–8.
[23] Akhoondi F, Lotfi MM. A heuristic algorithm for master production scheduling [44] Shrouf F, Ordieres-Meré J, García-Sánchez A, Ortega-Mier M. Optimizing the pro-
problem with controllable processing times and scenario-based demands. Int J Prod duction scheduling of a single machine to minimize total energy consumption costs.
Res 2016;54(12):3659–76. J Clean Prod 2014;67(1):197–207.

315