Professional Documents
Culture Documents
a r t i c l e i n f o a b s t r a c t
Keywords: Production scheduling under time-of-use electricity tariffs has become an efficient way for energy-intensive indus-
Scheduling tries to decrease energy costs. However, when production tasks are over-concentrated in one scheduling cycle,
Self-generation the effectiveness of time-of-use electricity tariffs is no longer significant. This makes the introduction of self-
Time-of-use electricity tariffs
generation power plant appealing for energy-intensive industries. This paper addresses an integrated scheduling
Multi-objective mathematical model
problem from an iron-steel plant equipped with self-generation equipment under time-of-use electricity tariffs. In
Relationship propagation chain
The improved SPEA2 this problem, time-of-use electricity tariffs, the self-generation cost, and the on-grid electrovalence influence the
total electricity cost simultaneously. A multi-objective mathematical model with energy-awareness is developed
to optimize the production schedules and electricity cost jointly. An improved SPEA2 based on the relationship
propagation chain is tailored for the problem, including scheduling solution encoding, crossover and mutation. A
real-life case study from a Chinese iron-steel plant equipped with self-generation equipment demonstrates that the
proposed methods can provide a high-quality scheduling scheme and the total electricity cost can be significantly
reduced.
1. Introduction to other periods, electricity providers have to start their backup power
facilities to meet customer demands, which results in prohibitive costs
Under the pressures of climate change, rising energy costs, and grow- [6]. A viable alternative to peak generation capacity is to involve con-
ing energy security concerns, researches on production scheduling have sumers in power balancing and use their flexibility to avoid the peaks.
increasingly been conducted from the points of view of energy-saving Techniques involving such practices are referred as Demand-Side Man-
and environmental protection. Electricity is one of the main energy agement (DSM) [7]. As one of the most commonly used strategies in
sources used in industrial production [1]. Therefore, the rising prices DSM, time-dependent electricity pricing has been widely implemented
of electricity have prompted many industrial enterprises to focus on in many countries. At the core of this strategy are the so-called time-of-
the issue of reducing electricity cost. Until recently, most efforts aim- use (TOU) electricity tariffs, under which electricity price varies from
ing at reducing the total electricity cost have focused on developing hour to hour throughout a day, depending on consumer demand. To
more energy-efficient machines and equipment [2]. Experimentations encourage consumers to consume more energy during non-peak peri-
and practices show that replacement of new and more advanced equip- ods, electricity prices in these periods are set below those in other pe-
ment requires substantial capital investment. Alternatively, some re- riods. Exploring scheduling problems under this kind of pricing scheme
searchers have demonstrated that electricity cost can be significantly provides a massive opportunity to reduce cost for electricity-intensive
reduced through employing operational strategies (such as lean process consumers by shifting electricity consumption from on-peak periods to
control) and optimizing production schedules [3–5]. off-peak or mid-peak periods.
Many production tasks and complex technological conditions lead to The effectiveness of TOU electricity tariffs is, however, no longer
the uneven demands for electricity in manufacturing industries through- significant when production tasks are over-concentrated in one sched-
out a day. According to the load variation of the power grid, a day can ule cycle [8]. This makes the introduction of self-generation power
be divided into several periods, such as on-peak, mid-peak, and off-peak plant appealing for energy-intensive consumers because it can be a
periods. Although the duration of on-peak periods is often short relative better strategy to reduce energy cost, except for price-based schemes.
∗
Corresponding author at: School of Management Science and Engineering, Anhui University of Technology, Ma’anshan, PR China.
E-mail address: alltimefight@gmail.com (R. Pan).
https://doi.org/10.1016/j.swevo.2020.100764
Received 2 October 2018; Received in revised form 25 December 2019; Accepted 18 August 2020
Available online 27 August 2020
2210-6502/© 2020 Elsevier B.V. All rights reserved.
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
At present, most manufacturing enterprises are equipped with self- ars [14–16]. Generally, scheduling problems without considering TOU
generation equipment. Self-generation cost arises from the use of gener- electricity tariffs in manufacturing tend to assume that the electricity
ation equipment to meet the in-plant demand for electricity. When the prices at different periods are the same. Consequently, decision objec-
amount of self-generation exceeds the in-plant demand, the plant can tives usually focus on time-related criteria, such as the minimization of
sell the surplus electricity to the electricity providers at a certain price makespan, total completion time, cost of tardiness, and earliness penal-
called the on-grid electrovalence, thus obtaining a certain profit. Other- ties for jobs. In contrast to the former problem, the decision objectives
wise, the plant will purchase power from the electricity providers and of scheduling problems under TOU electricity tariffs focus on reducing
the price of outsourcing electricity is consistent with the TOU electric- electricity cost, except for time-related criteria.
ity tariffs. It is possible that three types of electricity prices (i.e., TOU Recently, many researchers have investigated single machine and
electricity tariffs, self-generation cost, and on-grid electrovalence) can parallel machine scheduling problems under TOU electricity tariffs. For
sometimes exist simultaneously. Consequently, this complicates the cal- single-machine scheduling problems, Rubaiee and Yildirim [17] studied
culation of the total electricity cost, which sums the arithmetic product a preemptive scheduling problem to minimize the total completion time
of electric power consumption and the corresponding price in each time and total energy cost under TOU electricity tariffs. Several new holis-
interval. This presents an interesting challenge in terms of minimizing tic ant colony optimization algorithms were developed to solve it. Bi-
the total electricity cost of manufacturing industries equipped with self- objective single-machine batch scheduling under TOU electricity prices
generation equipment under TOU electricity tariffs. was addressed by Zhang et al. [18] to minimize the total energy cost
In this work, we explore the production scheduling problem of an and makespan. Wang et al. [19] investigated a single-machine batch
iron-steel plant equipped with self-generation equipment under TOU scheduling problem to minimize the makespan and total energy cost
electricity tariffs. The goal of this work is to find an optimal produc- simultaneously. They also developed two heuristic methods to obtain
tion schedule for an energy-intensive process plant to realize its demand approximate Pareto fronts based on decomposition ideas. Fang et al.
side response potential at the production scheduling level. The key of [20] examined a single machine scheduling problem to minimize the
the problem is to assign a set of jobs to available machines and time pe- total electricity cost of processing jobs under TOU electricity tariffs. For
riods with varying electricity prices, and then to determine the amount both the uniform-speed and speed-scaling cases, they showed how to
of self-generation. Based on the process constraints, the generator con- compute optimal schedules for the preemptive version of the problem in
straints and three types of electricity prices, a multi-objective mathemat- polynomial time. Cheng et al. [21] investigated a single-machine batch
ical model is proposed to minimize the makespan and total electricity scheduling problem under TOU electricity tariffs and they formulated a
cost at the same time. To enhance the effectiveness and efficiency of bi-objective MILP model to minimize the makespan and total electric-
SPEA2, we also propose the improvement strategies for SPEA2 based ity cost. An exact 𝜀-constraint method was adapted to obtain its Pareto
on the relationship propagation chain. To investigate and measure the front.
performance of the proposed algorithms, a real-life case is studied from The parallel machine scheduling problem is of great significance be-
an iron-steel plant in China. cause it is not only a generalization of the single-machine scheduling
The rest of this paper is structured as follows. Section 2 presents problem but it is also a special case of the flexible flow shop scheduling
the background and previous work on energy-aware scheduling prob- problem [22]. For parallel machine scheduling under the TOU electricity
lems. Section 3 is devoted to the problem description and a multi- tariffs, Cheng et al. [23] provided an improved MILP model by signifi-
objective mathematical model of the scheduling problem is proposed. In cantly reducing the number of decision variables. Moon et al. [24] inves-
Section 4, an efficient improved SPEA2 based on the relationship propa- tigated an unrelated parallel machine scheduling problem to minimize
gation chain is developed. In Section 5, the performance of the improved the weighted sum of makespan and electricity cost. They proposed a
SPEA2 is tested through a real-life instance from the iron-steel industry hybrid genetic algorithm with a blank job insertion algorithm to solve
in China. Finally, conclusions and future research directions are given the problem. Che et al. [25] investigated an energy-conscious unrelated
in Section 6. parallel machine scheduling problem under TOU electricity tariffs. They
first built an improved continuous-time MILP model for the problem and
2. Literature review they then proposed a two-stage heuristic algorithm. An unrelated par-
allel machine scheduling problem under TOU electricity tariffs to mini-
Several avenues have been pursued when scheduling to reduce to- mize the total electricity cost while ensuring that the overall completion
tal energy consumption. In [9], an energy-efficient hybrid flow shop time does not exceed a predetermined production deadline was stud-
scheduling problem with total tardiness, makespan and total energy con- ied by Ding et al. [26]. They modeled a new time-interval-based MILP
sumption was addressed. A new Pareto dominance was defined to deal formulation and then reformulated the problem using a Dantzig–Wolfe
with objectives relative importance and a two-level imperialist competi- decomposition.
tive algorithm was presented. Aghelinejad et al. [10] investigated a non- On the basis of researches on single-machine and parallel-machine
preemptive single-machine manufacturing environment to reduce total energy-aware scheduling, some other researchers have explored the
energy costs. New mathematical models with the production scheduling shop scheduling problem under TOU electricity tariffs. Wang et al.
at machine level as well as job level are presented. Lu et al. [11] inves- [27] investigated a bi-criteria energy-efficient two-stage hybrid flow
tigated an energy-efficient permutation flow shop scheduling problem shop scheduling problem with consideration of makespan and total en-
with sequence-dependent setup and controllable transportation time, ergy consumption. TOU electricity tariffs and different states of ma-
and proposed a hybrid multi-objective backtracking search algorithm chines (working, idle and turnoff) were integrated. Zheng et al. [28] pre-
to solve it. Liu et al. [12] introduced a model for the bi-objective op- sented a multi-objective hybrid ant colony optimization algorithm for a
timization problem that minimizes the total non-processing electricity real-world two-stage blocking permutation flow shop scheduling prob-
consumption and total weighted tardiness in a job shop. A novel multi- lem to address the trade-off between total energy costs and makespan
objective genetic algorithm based on NSGA-II was developed. Mansouri as measures of the service level with TOU electricity price.
and Aktas [13] developed a multi-objective genetic algorithm for a two- A few studies have focused on the production scheduling problem
machine sequence-dependent permutation flowshop scheduling prob- for the iron-steel manufacturing under TOU electricity tariffs. Yang et al.
lem to address the trade-off between sustainability and service level. [29] integrally studied the problems of scrap steel charge optimization
The above literatures focused on reducing electricity consumption, but considering metal elements concentrations uncertainty and scrap steel
not considered electricity pricing policies. melt shop production scheduling optimization using robust optimiza-
To achieve the goal of sustainable manufacturing, scheduling prob- tion approaches. Hadera et al. [30] developed a MILP formulation to
lems under the TOU electricity tariffs have been studied by many schol- extend a continuous-time model with energy-awareness to optimize the
2
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
(1) All of the charges are grouped into several casts according to steel
grade, the properties of the blanks, the customers’ orders, and the
life of tundish (which is a critical component of the CCM). Given
that the cast planning problem has been studied in depth, we as-
sume that the cast sequences on the casters and charge sequences
in each cast are predetermined.
(2) All of the charges in the same cast must be processed consecu-
tively on a same CCM.
(3) The set-up time must be considered for the first charge in each
Fig. 1. Flow diagram of the steelmaking process. cast at the continuous casting stage.
3
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
𝑝𝑚𝑎𝑥
𝑙,𝐿
: the maximum processing time of charge j in cast i at stage l; 𝐿𝑘 = 𝑒𝑘∕(𝑡 − 𝑡 ), 𝑘 = 1, ..., 𝐾 (15)
𝑖𝑗 𝑘 𝑘−1
E: the set of the precedence relationship of casts, (i1 , i2 )∈E, implying
that casts i1 and i2 are processed on the same CCM and i1 is the 𝑄𝑘 = 𝐿𝑘 − 𝑃𝑘 , 𝑘 = 1, … , 𝐾, (16)
directly predecessor of i2 ;
{
S: the set-up time of CCM between two adjacent casts; (𝑇𝑘 − 𝑇𝑘−1 )𝑝𝑜𝑢𝑡
𝑘
𝑄𝑘 , 𝑄𝑘 > 0
esti : the earliest starting time of cast i; 𝐵𝑘 = , 𝑘 = 1, … , 𝐾, (17)
lsti : the latest starting time of cast i; (𝑇𝑘 − 𝑇𝑘−1 )𝑝𝑠𝑒𝑙𝑙
𝑘
𝑄𝑘 , 𝑄𝑘 ≤ 0
ml : the total number of machines at production stage l;
𝑃 min ≤ 𝑃𝑘 ≤ 𝑃 max , 𝑘 = 1, … , 𝐾, (18)
dl : the unit energy consumption of machine at production stage l;
UB: the upper bound of the scheduling cycle, 1440 min;
|𝑃 − 𝑃 | ≤ Δ𝑃 , 𝑘 = 1, … , 𝐾, (19)
k: index of time intervals, k =1,..., K, where K is the total number of | 𝑘 𝑘−1 |
time intervals in the involved scheduling cycle;
Tk -1 : the starting time in time interval k, where TK = UB; 𝑝𝑚𝑖𝑛 𝑚𝑎𝑥
𝑙,𝐿 ≤ 𝑝𝑙,𝐿𝑖𝑗 ≤ 𝑝𝑙,𝐿 , 𝑙 = 1, ..., 𝑙𝑚𝑎𝑥 , 𝑖 = 1, ..., 𝐺, 𝑗 ∈ Ω𝑖 , (20)
𝑖𝑗 𝑖𝑗
Pk gen : the self-generation cost in time interval k;
Pk out : the outsourcing electricity price in time interval k (i.e., TOU 0 ≤ 𝑠𝑙,𝐿𝑖𝑗 ≤ 𝑈 𝐵, 𝑙 = 1, ..., 𝑙𝑚𝑎𝑥 , 𝑖 = 1, ..., 𝐺, 𝑗 ∈ Ω𝑖 , (21)
electricity tariffs);
Pk sell : the on-grid electrovalence in time interval k;
𝑒𝑠𝑡𝑖 ≤ 𝑠𝑖 ≤ 𝑙𝑠𝑡𝑖 , 𝑖 = 1, ..., 𝐺, (22)
Pmax : the maximum power output of the generator;
Pmin : the minimum power output of the generator; The objective function given in Eq. (1) minimizes the sum of the
∆P: the ramp-up or ramp-down rate of the generator; makespan of all casts. The objective function given in Eq. (2) mini-
Decision variables: mizes the total electricity cost of processing all casts. Eq. (3) guaran-
𝑝𝑙,𝐿𝑖𝑗 : the processing time of charge j in cast i at stage l; tees that the waiting time of two adjacent production stage of a same
𝑠𝑙,𝐿𝑖𝑗 : the relative starting time of charge j in cast i at stage l; charge is larger than the transportation time but not exceeding the
si : the starting time of cast i; maximum allowable limit. Eq. (4) ensures the continuous casting con-
Pk : the amount of self-generation in time interval k. straint. Eq. (5) specifies the precedence relationship of adjacent charges.
Eq. (6) defines the waiting time. Eq. (7) defines the relative starting time
The following programming model of the multi-casts scheduling
of cast i at production stage l. Eq. (8) defines the relative ending time of
problem with consideration of self-generation under TOU electricity tar-
cast i at production stage. Eq. (9) ensures the precedence relationship for
iffs is established as follows.
casts processed on a same CCM. Eq. (10) defines the constraint of ma-
𝑁
∑ ( ) chines number. At any time, the required machine’s number should not
𝑚𝑖𝑛 𝑚𝑎𝑥 𝑠𝑙𝑚𝑎𝑥 ,𝐿𝑖𝑗 + 𝑝𝑙𝑚𝑎𝑥 ,𝐿𝑖𝑗 (1)
exceed the maximum availability provided by each production stage.
𝑖=1
∑ ∑ Eqs. (11)–(14) are used to calculate the total energy consumption of all
𝑚𝑖𝑛 𝐵𝑘 + (𝑇𝑘 − 𝑇𝑘−1 )𝑝𝑔𝑒𝑛
𝑘
𝑃𝑘 (2) casts. Eq. (11) calculates the total energy consumption for each cast in
𝑘=1 𝑘=1
s.t. time interval k. Eq. (12) computes the energy consumption for each op-
eration in time interval k. The overlap of processing time of a cast in
𝑡𝑟𝑙 ≤ 𝑤𝑙,𝐿𝑖𝑗 ≤ 𝑤𝑡, 𝑙 = 1, ..., 𝑙𝑚𝑎𝑥 , 𝑖 = 1, ..., 𝐺, 𝑗 ∈ Ω𝑖 , (3) time interval k is calculated based on the method used in Nolde and
( ) ( ) Morari [34]. Eq. (13) defines the processing time of a cast in time inter-
𝑠𝑙𝑚𝑎𝑥 ,𝐿𝑖𝑗 − 𝑠𝑙𝑚𝑎𝑥 ,𝐿𝑖𝑗 + 𝑝𝑙𝑚𝑎𝑥 ,𝐿𝑖𝑗 = 0, 𝑖 = 1, ..., 𝐺, 𝐿𝑖𝑗1 , 𝐿𝑖𝑗2 ∈ 𝐷, (4) val k, when the completion time is prior to Tk . The duration for a cast at
2 1 1
each production stage can be calculated based on the schedule plan for
( ) ( ) charges. Eq. (14) defines the processing time of a cast in time interval
𝑠𝑙,𝐿𝑖𝑗 − 𝑠𝑙,𝐿𝑖𝑗 + 𝑝𝑙,𝐿𝑖𝑗 ≥ 0, 𝑙 = 1, ..., 𝑙𝑚𝑎𝑥−1 , 𝑖 = 1, ..., 𝐺, 𝐿𝑖𝑗1 , 𝐿𝑖𝑗2 ∈ 𝐷, k, when the starting time is later than Tk -1 . In particular, Eqs. (15)–(19)
2 1 1
(5) are the extra constraints to be considered when an iron-steel plant is
equipped with self-generation equipment. Eq. (15) calculates the total
( ) electric demands in time interval k. Eq. (16) computes the difference
𝑤𝑙,𝐿𝑖𝑗 = 𝑠𝑙+1,𝐿𝑖𝑗 − 𝑠𝑙,𝐿𝑖𝑗 + 𝑝𝑙,𝐿𝑖𝑗 , 𝑙 = 1, ..., 𝑙𝑚𝑎𝑥 , 𝑖 = 1, ..., 𝐺, 𝑗 ∈ Ω𝑖 , (6)
between the amount of self-generation and the total electric demands
in time interval k. Eq. (17) calculates the earnings of selling electricity
𝑡𝑙 ,𝑖 = 𝑚𝑖𝑛𝑠𝑙,𝐿𝑖𝑗 , 𝑖 = 1, ..., 𝐺, 𝑗 ∈ Ω𝑖 , (7) to electricity providers or outsourcing electricity cost in time interval k.
( ) Eq. (18) defines the range of output of the generator and Eq. (19) defines
𝑡𝑙𝑚𝑎𝑥+𝑙 ,𝑖 = 𝑚𝑎𝑥 𝑠𝑙,𝐿𝑖𝑗 + 𝑝𝑙,𝐿𝑖𝑗 , 𝑖 = 1, ..., 𝐺, 𝑗 ∈ Ω𝑖 , (8) the ramp-up or rate-down rate of the generator. Finally, Eqs. (20)–(22)
define the range of decision variables.
( ) ( )
𝑠𝑖2 + 𝑡𝑙𝑚𝑎𝑥 , 𝑖2 − 𝑠𝑖1 + 𝑡𝑙𝑚𝑎𝑥 ×2,𝑖1 ≥ 𝑆, (𝑖1, 𝑖2) ∈ 𝐸, 𝑖1, 𝑖2 = 1, … , 𝐺, (9)
4
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
5
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
conflict position of 𝛿3 is eliminated. We then end the adjustment, and ferent adjustment strategies according to the relationship between the
update 𝑝3,𝑗1 (38=37+1) and the corresponding scheduling solution of T starting time of the current charge and the ending time of the preced-
and W after adjusting 𝛿3, as shown in Fig. 5. The values after adjusting ing charge at a same production stage. Suppose that the conflict posi-
𝑝3,𝑗1 and 𝑤2,𝑗2 are marked in red. tion is charge j at production stage l, for the case of sl, j is lower than
We can see from Fig. 5 that 𝑠3,𝑗3 is larger than 𝑠3,𝑗2 + 𝑝3,𝑗2 , which is 𝑠𝑙,𝑗−1 + 𝑝𝑙,𝑗−1 , we first lessen 𝑝𝑙,𝑗−1 and we then enlarge 𝑤𝑙−1,𝑗 , if neces-
similar to the case of conflict position 𝛿3, while 𝛿2 is still in the conflict sary. Where sl, j is larger than 𝑠𝑙,𝑗−1 + 𝑝𝑙,𝑗−1 , we first lessen 𝑤𝑙−1,𝑗 and we
position. Similarly, we adjust 𝛿2 according the adjusting method of 𝛿3. then enlarge 𝑝𝑙,𝑗−1 , if necessary.
We decrease 𝑤2,𝑗3 at the first step; that is, the value 19 between 43(3, 2) For the first case, Fig. 8 shows our adjustment strategy in detail. Sup-
and 51(3, 3) in Fig. 5. The maximum allowable adjustment value 𝜎2(11) pose the conflict position 𝛿 is charge j at production stage l, and charge
is the difference between 𝑤2,𝑗3 (19) and tr2 (7) in Table 1. Since 𝜎2 is j-1 is the preceding charge of charge j. The minimum value that needs to
larger than the minimum adjusted value 8 in 𝛿3, the conflict position 𝛿3 be adjusted in 𝛿 is𝑉𝛿 (𝑠𝑙,𝑗−1 + 𝑝𝑙,𝑗−1 − 𝑠𝑙,𝑗 ). The process time of charge j-1
is eliminated. We end the adjustment and we then update 𝑤2,𝑗3 , which at l production stage is equal to the element T(j-1,l) in matrix T, where
is marked in red (11=19-8), and the corresponding scheduling solution 𝑝min is the minimum process time of charge j-1 at the production stage l.
𝑙,𝑗−1
of T and W after adjusting 𝛿2, as shown in Fig. 6. The waiting time of the current charge between the production stage l-1
After individually adjusting 𝛿1, 𝛿3, and 𝛿2, all of the conflict posi- and l is equal to the element W(j,l-1) in matrix W, and wt is the maximum
tions are eliminated, and the scheduling solution of T and W becomes waiting time between production stage l-1 and l. In the first step, we de-
feasible. Matrixes T and W are shown in Fig. 7, in which the adjusted crease T(j-1,l) based on Eqs. (4), (5) and (20). If the maximum allowable
values are marked with an ∗ . adjustment value 𝜎1 of T(j-1,l) is larger than V𝛿 , then 𝛿 is eliminated.
We propose a general two-step adjustment strategy according to the We end the adjustment and update T(j-1, l) and V𝛿 ; otherwise, we need
adjustment procedure for conflict positions in the infeasible T and W of to further increase W(j,l-1) based on Eq. (3). If the maximum allowable
a cast. We divide the conflict positions into two cases, which adopt dif- adjustment value 𝜎2 of W(j,l-1) is larger than V𝛿 , then 𝛿 is eliminated
6
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
then 𝛿 is eliminated, and we end the adjustment and update T(j-1, l);
otherwise, we quit the adjustment.
For randomly generated T and W at each cast, we evaluate their
feasibility by the scheduling solution. If the scheduling solution is infea-
sible, we first find out the conflict positions, then judge which conflict
case it belongs to, and select corresponding adjustment strategy. Since
the allowable adjustment of process time is large for all of charges, the
Fig. 7. T and W for the cast contained 3 charges after adjustment.
proposed approach can usually transform the infeasible T and W into
the feasible ones within two steps.
7
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Input: The population size, NP; range of processing time at stage l, tjl ;
transportation time of cast j at stage l, tr slj ; The machine set at stage l, ml
Output: The initial population, P0 ={𝑋10 , 𝑋20 , ...𝑋𝑖0 ..., 𝑋𝑁𝑃
0
}
1. initialize P0
2. set feasibilityTest = 0
3. generate an initial head node T and node W for all casts // Eqs. (3) and
(20)
4. for i =1 to NP do
5. while feasibilityTest == 0 do
6. Xi .RST← Xi .(T,W)
7. Xi .RET←Xi .(T,W) // Eqs. (6)–(8)
8. Randomly generate the inner node R.
9. Xi .M←Xi .(R,ml )
10. Xi .ST←Xi .(R,M,RST) // Eqs. (22)
11. Xi .ET←Xi .ST + Xi .RET
12. feasibilityTest () // Eqs. (4)and(5)
13. end while
14. X.P←Cost(X.ST, X.ET)// Eqs. (18) and (19)
15. i ← i+1
16. end for
Fig. 8. The two-step adjustment strategy for case of 𝛿1. 17. return P0 {T, W, R, M, RST, RET, ST, ET, P}
8
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Algorithm 2 Hybrid crossover operator based on the concentration-diffusion Algorithm 3 Mutation operator.
mechanism.
Input: The population 𝑃 = {𝑋1 , 𝑋2 , ⋯ 𝑋𝑖 ⋯ , 𝑋𝑁𝑃 }; mutation probability, pm
Input: The population 𝑃 = {𝑋1 , 𝑋2 , ⋯ 𝑋𝑖 ⋯ , 𝑋𝑁𝑃 }; crossover probability, pc 1 ; Output: The mutated population P’
Output: The crossed population P’ set P’= P
set P’ = P set feasibilityTest = 0
for i =1 to NP do for i =1 to NP do
Calculate Xi .rank and Xi .dist while feasibilityTest == 0 do
end for Randomly choose two positions, j1 and j2
set y = 1 𝛾= random(0, Δt)
set feasibilityTest = 0 if rand ≤ pm
for i =1 to NP/2 do max(T(j1 , l), T (j2 , l))←max(max(T(j1 , l), T(j2 , l)) - 𝛾, min(T(j1 , l), T (j2, l))+ 𝛾)
while feasibilityTest == 0 do min(T(j1 , l), T(j2 , l))←min(max(T(j1 , l), T(j2 , l)) - 𝛾, min(T(j1 , l), T(j2 , l))+ 𝛾)
Randomly choose two chromosomes1 and 2 RST ← (T′, W′)
Δt= |T1- T2| RET ← (T′, W′)
rc2 = random(0, Δt) Randomly generate the inner node R.
if rand <pc 1 M ← (R, ml )
T1←max(T1, T2) ST ← (R, M, RST)
T2←min(T1, T2) ET ← ST+RET
W1←max(W1, W2) end if
W2←min(W1, W2) feasibilityTest ()
else end while
max(T1, T2)←max(max(T1, T2)- rc2 , min(T1, T2) + rc2 ) P←Cost(ST, ET)
min(T1, T2)←min(max(T1, T2)- rc2 , min(T1, T2) + rc2 ) i ← i+1
(RST1 , RST2 )←(T1 , W1 , T2 , W2 ) end for
(RET1 , RET2 )←(T1 , W1 , T2 , W2 ) return P’{T, W, R, M, RST, RET, ST, ET, P}
Randomly generate the inner node R1 , R2 .
( M 1 , M 2 )← ( R 1 , m l 1 , R 2 , m l 2 )
(ST1 , ST2 )← (R1 , M1 , RST1 , R2 , M2 , RST2 ) Table 3
(ET1 , ET2 )← (ST1 +RET1 , ST2 +RET2 ) The ranges of the minimum and maximum processing time in
end if each stage.
feasibilityTest () // Eq. (30)-(32)
end while l=1 l=2 l=3
P1←Cost (ST1, ET1) min max min max
pl pl pl pl pl min pl max
P2←Cost (ST2, ET2)
[35,40] [50,55] [15,20] [55,60] [35,40] [50,55]
i ← i+1
end for
return P’{T, W, R, M, RST, RET, ST, ET, P}
worst case. Therefore, the whole complexity of the improved algorithm
is O((GK+max{Ni })lmax ).
smaller value 43 plus rc2 (43+2=45). We then assign 50 to T1 and 45 to
T2. Similarly, we do the same conversion for all of elements in matrixes 5. Computational experiments and results
T1, T2, W1 and W2. The crossover results are shown in Fig. 12, where
the changed values are marked in red. Our algorithm was implemented in MATLAB 2015b on an Intel Core
i7-4940 3.6-GHz using the Linux operating system. To evaluate the per-
4.3. Mutation operator formance of the proposed heuristic algorithm, we used the same test
instances to make a fair comparison. The parameter settings that we
In SPEA2, the mutation operator helps to impose variety and diver- used are as follows. The transportation time between the steelmaking
sity in the population. Since there are many constraints of the scheduling and refining stage is randomly generated from a uniform distribution
problem, the traditional mutation operator is impracticable because it (2, 5) minutes, while the transportation time between the refining and
easily falls into the local optimum. Therefore, we propose a powerful casting stage is randomly generated in (6, 10) minutes. The unit energy
mutation operator, which is accomplished through the following steps consumption of EAFs is d1=1000 KWH, for LFs d2=150 KWH and CCMs
in Algorithm 3, and we give an example in Example 4. d3=50 KWH. The ranges of the minimum and maximum processing time
for charges in each stage are illustrated in Table 3. The maximum wait-
Example 4. We randomly choose one chromosome from the cur- ing time is W = 20 min and the set-up time is S = 5 min. The output of
rent population. According to Algorithm 3, two positions (j1 =1 and the generator ranges from 95 to 150 MW, and the ramp-up/down rate
j2 =2) are randomly generated. Next, we take the elements T(1,1)(43) of the generator is 40 MW/h. We use three types of electricity prices of
and T(2,1)(42) as an example, and we then calculate the difference TOU electricity tariffs, self-generation cost, and on-grid electrovalence,
Δt(Δt=|43- 42|=1) and randomly generate a value (𝛾=1) between 0 and which are implemented in Shaanxi Province, China, for general industry
Δt. We select the bigger value 43 minus 1 (43-1=42), the smaller value (under 10 KV), as given in Table 4. It is assumed that 0 a.m. of one day
42 plus 1 (42+1=43). Since T(1,1) is bigger than T(2,1), 43 and 42 are is the start point of the scheduling horizon.
assigned to T(1,1) and T(2,1), respectively. Similarly, we do the same
conversion for all of elements in matrixes T and W. The mutation result 5.1. Performance analysis of the model and heuristic algorithm
is shown in Fig. 13, where the changed values are marked in red.
In this section, we apply the improved SPEA2 to solve the scheduling
4.4. Complexity of the improved SPEA2 based on the hybrid adjustment problem of multi-casts. Fig. 14 shows the Pareto fronts of four casts’ case.
strategy of the single cast stage To evaluate the performance of the improved algorithm, we first com-
pare the makespan with CPLEX, which was used to solve the scheduling
Given that there are G casts, lmax production stages, K periods of problem of single cast in Tan and Liu [31]. The makespan of four single
time, and Ni charges in single cast i, the complexity of the above- casts obtained by CPLEX are 2028 min, while the minimum makespan
mentioned encoding, crossover, and mutation operator for SPEA2 are obtained by our approach with consideration of multi-casts simultane-
O((GK+max{Ni })lmax ), O((GK+max{Ni })lmax ), and O(GKlmax ) in the ously can reach 2010 min.
9
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Fig. 11. The crossover results for T1, W1, T2, W2 for
the chosen cast.
Second, we compare the total electricity cost between the iron-steel reveal that the total electricity cost can be reduced by 27.11% at most
plant under TOU electricity tariffs and the iron-steel plant equipped with when the iron-steel plant is equipped with self-generation equipment.
self-generation equipment under TOU electricity tariffs. The total elec- Each solution in the Pareto fronts corresponds to a scheduling
tricity cost under TOU electricity tariffs is 1 054 376 yuan. However, the scheme. We show the scheduling scheme of the first solution in the
total electricity cost reduced observably in our results. We can see from Pareto fronts in Figs. 15 and 16. The box with the same color in
the Fig. 14, the maximum total electricity cost is 825 385 yuan, and the Fig. 15 represents a cast. The value in the box represents the index of
minimum total electricity cost can be down to 768 785 yuan. The results cast and the production stage it lies. For example, (1, 1) represents Cast1
10
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Fig. 12. The crossover results for T1, W1, T2, W2 for the chosen cast.
Table 4
Three types of electricity prices.
Period type TOU electricity tariffs (CNY/KWH) Self-generation cost (CNY/ KWH) On-grid electrovalence (CNY/ KWH) Time periods
11
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Fig. 16. The scheduling solution for all charges included in the same cast.
at the first stage. We can see that the scheduling solution meets the con- shown in Fig. 16. Each box represents a charge with its production
straints of the number of machines and precedence relationship for casts stage. The value in a box represents the process time of a charge at
processed on the same CCM. Cast1 and Cast2 are at the same CCM, and the current production stage. We can see that the scheduling solu-
Cast1 is prior to Cast2. Cast3 and Cast4 are at the same CCM, and Cast3 tion meets the process constraints of the waiting time, transportation
is prior to Cast4. time of two adjacent production stages, and continuous casting con-
The four casts in Fig. 15 have different number of charges and straint. The results of the experiment have confirmed that the proposed
the scheduling solutions for all charges included in same cast are approach is effective to produce the daily schedule for an iron-steel
12
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Table 5
Performance of the improved algorithm and CPLEX.
13
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Table 6
Mean and standard deviation of ∆ and IGD with iterations of 20, 25, 30, 35 and 40.
iterations
metric
20 25 30 35 40
Table 7
Results of six algorithms on IGD (Case 1).
Table 8
Results of six algorithms on Spread (Case 1).
20 0.811 0.004 0.873 0.018 0.847 0.008 0.9774 0.008 1.0804 0.0119 1.1629 0.0147
25 0.795 0.008 1.035 0.023 1.006 0.007 0.9405 0.007 1.0431 0.0134 1.1524 0.0177
30 0.787 0.003 1.300 0.022 1.160 0.002 0.9095 0.019 1.0259 0.0122 1.1338 0.0128
35 0.700 0.017 1.159 0.028 1.115 0.003 0.8653 0.013 1.0494 0.0178 1.1438 0.0103
40 0.674 0.026 1.102 0.021 1.122 0.003 0.9719 0.014 1.0623 0.0105 1.1538 0.0175
Due to the historical information guiding the evolution of population, but still less diverse than the improved SPEA2 (the values of mean are a
HMOBSA can produce populations with significant differences in the bit larger). The mean and var of the improved SPEA2 are much smaller
early evolution stage, which has strong global exploration ability; how- than those of NSGA-II and NSGA-III, which indicates that the diversity
ever, in the later evolution stage, the difference between the new popu- of the improved SPEA2 is better and its performance is more stable.
lation and historical population is small, which is not conducive to the MOACO and MODE show better population diversity and stability than
further exploration of the algorithm in the solution space; therefore, it HMOBSA. The above results can also be verified from Fig. 17.
is prone to premature. In solving our problem, MOACO and MODE are Tables 9 and 10 describe the IGD values and running time for dif-
prone to local optimization and present poor convergence, because the ferent cases under iterations = 40, respectively. We can observe that:
performance of them is highly dependent on control parameters which (1) considering the IGD values, the proposed algorithm obtains optimal
need to determine by numerous experiments. values for all of cases except Case 2, which indicates our algorithm is
superior to other comparison algorithms in convergence and diversity;
5.2.2. The superiority of the improved SPEA2 in evaluation metrics (2) comparing running times, our algorithm is faster than the others in
Ten independent simulation runs are conducted using six algorithms most cases. Although it is not optimal in a few cases, it has a small dif-
respectively with different iterations of 20, 25, 30, 35 and 40. Tables 6– ference with the optimal time, which verifies the time efficiency of the
8 take Case 1 as an example to record the change process of evaluation algorithm.
metrics as the number of iterations increases. Tables 9 and 10 report the
computational results and running time of them in different cases. 5.3. Energy consumption curve
Table 6 shows the mean and standard deviation values of ∆ and
IGD about the improved SPEA2. We can see that, ∆ becomes smaller as We have shown the feasibility of the scheduling solution by ana-
iteration increases while IGD remains stable relatively under different lyzing the process constraints from the previous section, and we then
iterations. Since both of them are relatively small in mean and standard show the superiority of the scheduling solution by analyzing the energy
deviation, it reveals that the Pareto front obtained by our algorithm has consumption curve. The energy consumption curve clearly reflects the
a good spread and converges to the true Pareto-optimal front. electric power consumption distribution in one schedule cycle. It also
The IGD and Spread values among six algorithms under itera- provides an effective way to analyze the performance of different opti-
tions = 20, 25, 30, 35 and 40 are reported in Tables 7 and 8, respectively. mal algorithms.
It can be seen from Table 7 that the IGD value of our algorithm has been The electric power consumption curve and amount of self-generation
in a stable state, because the algorithm has been approximately con- of four casts are shown in Fig. 18. EPC represents the electric power
verged in about 20 generations. The IGD values of NSGA-II and NSGA-III consumption in each time interval while ASG represents the amount of
gradually stabilized around 30 generations. But their values are bigger self-generation in each time interval, and the blue histogram graph de-
than those of the improved SPEA2, which means Pareto Fronts of them notes the TOU electricity tariffs. Although ASG is in accordance with the
are dominated by the improved SPEA2. The values of MOACO are the change of EPC, they are different in some time intervals. In time interval
worst and are constantly changing, which indicates that the convergence A, ASG is 0 KW because the self-generation cost is higher than the TOU
and diverse of the algorithm is poor. The Spread values among six al- electricity tariffs and EPC can be met by the electricity providers. In
gorithms are shown in Table 8. Clearly, NSGA-III is better than NSGA-II time interval B, ASG is larger than EPC because the self-generation cost
with great stability of diversity (the values of var are much smaller), is lower than the TOU electricity tariffs and the surplus power is sold to
14
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Table 9
Results of six algorithms on IGD (iterations 40).
Table 10
Computational times of six algorithms (iterations 40).
Running Time(s)
Case
the improved SPEA2 NSGA-III NSGA-II MOACO MODE HMOBSA
Fig. 18. The curves of electric power consumption and the amount of self-generation.
15
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
electricity providers. In time interval C, ASG is equal to EPC. It can be the concentration-diffusion mechanism and a new mutation approach
seen that the EPC of our scheduling solution is higher in on-peak and are proposed.
mid-peak periods for the iron-steel plant equipped with self-generation Experimental results have revealed that the improved algorithm is
under TOU electricity tariffs. This is different from the traditional sched- effective in finding high-quality solutions in reasonable time, meeting
ule solutions under TOU electricity tariffs, which decrease the energy the complicated and changeable cases of production scheduling, and
consumption in on-peak periods and increase the energy consumption the total electricity cost can decrease drastically with less total com-
in off-peak periods. Since EPC can mainly be met by the self-generation pletion time of jobs. In addition, we have two new findings. On the one
in on-peak and mid-peak periods and the self-generation cost is lower hand, the energy consumption of our scheduling solution is higher in on-
than the TOU electricity tariffs in two different periods, the electricity peak and mid-peak periods when the iron-steel plant is equipped with
cost is lower in these periods. self-generation equipment under TOU electricity tariffs. This is differ-
The blue line in Fig. 19 represents the TOU electricity tariffs, and ent from the traditional schedule solution under TOU electricity tariffs,
the red line represents the new electricity price that is generated by which shift electricity usage from on-peak periods to off-peak or mid-
TOU electricity tariffs, self-generation cost, and on-grid electrovalence peak periods. On the other hand, a new electricity price arises from TOU
combined. We can see that the new electricity price is always lower or electricity price, self-generation cost, and on-grid electrovalence. There-
equal to the TOU electricity tariffs. Therefore, the total electricity cost is fore, the total electricity cost can be formulated as the sum of the arith-
reduced, even though the total power consumption is higher in on-peak metic product of electric power consumption and new electricity price
periods. in each time interval. In our scheduling solutions, the new electricity
price is always lower or equal to the TOU electricity tariffs, which low-
ers the total electricity cost, even though the total power consumption
6. Conclusions is higher in on-peak periods.
In our future research, the machine restart mechanism of the steel-
This paper addresses a daily schedule of an iron-steel plant equipped making process under a new electricity price will be studied to decrease
with self-generation equipment under TOU electricity tariffs. The prob- the electricity cost because idle machines consume a large amount of
lem that we have considered is different from the traditional scheduling energy. In addition, the self-generation amount of iron-steel plant also
problem under TOU electricity tariffs because the electricity price is si- comes from the utilization of residual heat and energy in EAFs and heat-
multaneously influenced by TOU electricity tariffs, self-generation cost, ing furnaces, which are related to production loads. Therefore, to make
and on-grid electrovalence. The goal of the scheduling problem is to re- a more accurate scheme, it is recommended that the response approach
duce the makespan and total electricity cost by determining the optimal should be used to study the amount of self-generation, outsourcing, and
amount of self-generation and the optimal starting time of all of charges selling power in light of self-generation by EAFs and heating furnaces.
at each production stage. Based on the characteristics of the problem, a
multi-objective mathematical model is proposed. We conduct a relation-
ship propagation chain which is according to the propagation relation- Declaration of Competing Interest
ship of nine parts in a chromosome. Furthermore, to enhance the global
and local search ability of SPEA2, a hybrid crossover operator based on None
16
J. Cao, R. Pan, X. Xia et al. Swarm and Evolutionary Computation 60 (2021) 100764
Acknowledgements [16] H. Zhang, F. Zhao, K. Fang, et al., Energy-conscious flow shop scheduling under
time-of-use electricity tariffs, CIRP Ann. - Manuf. Technol. 63 (1) (2014) 37–40.
[17] S. Rubaiee, M.B. Yildirim, An energy-aware multiobjective ant colony algorithm to
This work is supported by the National Natural Science Foun- minimize total completion time and energy cost on a single-machine preemptive
dation of China (Grant Nos. 71772002 and 61702006) and the scheduling, Comput. Ind. Eng. 127 (2019) 240–252.
Open Fund of Key Laboratory of Anhui Higher Education Institutes [18] S. Zhang, A. Che, X. Wu, et al., Improved mixed-integer linear programming model
and heuristics for bi-objective single-machine batch scheduling with energy cost
(CS2019-ZD01). consideration, Eng. Optim. 50 (8) (2018) 1380–1394.
[19] S. Wang, M. Liu, F. Chu, et al., Bi-objective optimization of a single machine batch
Author Statement scheduling problem with energy cost consideration, J. Clean. Prod. 137 (2016)
1205–1215.
[20] K. Fang, N.A. Uhan, F. Zhao, et al., Scheduling on a single machine under time-of-use
Manuscript title: An efficient scheduling approach for an iron-steel electricity tariffs, Ann. Oper. Res. 238 (1-2) (2016) 199–227.
plant equipped with self-generation equipment under time-of-use elec- [21] J. Cheng, F. Chu, C. Chu, et al., Bi-objective optimization of single-machine batch
scheduling under time-of-use electricity prices, RAIRO – Oper. Res. 50 (4-5) (2016)
tricity tariffs
715–732.
Ruilin Pan contributed to the overall idea, algorithm, and wrote the [22] G. Weiss, in: Scheduling: Theory, Algorithms, and Systems by Michael Pinedo,
manuscript; Jianhua Cao, Xue Xia, and Xuemin Wang coded algorithms Springer, Berlin Heidelberg, 2016, pp. 35–42.
and contributed to the detailed writing as well as the revision, prepara- [23] J.H. Cheng, F. Chu, M.C. Zhou, An improved model for parallel machine schedul-
ing under time-of-use electricity price, IEEE Trans. Autom. Sci. Eng. 15 (2) (2018)
tion, and publishing of this paper; Xue Xia and Xuemei Shao contributed 896–899.
to data analyses and discussions on the scheduling model considering [24] J.Y. Moon, K. Shin, J. Park, Optimization of production scheduling with time-depen-
energy consumption; Xue Xia analyzed the characteristics of the prob- dent and machine-dependent electricity cost for industrial energy efficiency, Int. J.
Adv. Manuf. Tech. 68 (1-4) (2013) 523–535.
lem. All authors have read and approved the final manuscript. [25] A. Che, S. Zhang, X. Wu, Energy-conscious unrelated parallel machine scheduling
under time-of-use electricity tariffs, J. Clean. Prod. 156 (2017) 688–697.
Supplementary materials [26] J.Y. Ding, S. Song, R. Zhang, et al., Parallel machine scheduling under time-of-use
electricity prices: new models and optimization approaches, IEEE Trans. Autom. Sci.
Eng. 13 (2) (2016) 1138–1154.
Supplementary material associated with this article can be found, in [27] S.J. Wang, X.D. Wang, F. Chu, et al., An energy-efficient two-stage hybrid flow shop
the online version, at doi:10.1016/j.swevo.2020.100764. scheduling problem in a glass production, Int. J. Prod. Res. 1 (2019) 32.
[28] X. Zheng, S.C. Zhou, R. Xu, et al., Energy-efficient scheduling for multi-objective
two-stage flow shop using a hybrid ant colony optimisation algorithm, Int. J. Prod.
References
Res. 12 (2019) 1–18.
[29] Y. Yang, W.D. Chen, L. Wei, et al., Robust optimization for integrated scrap
[1] C.W. Park, K.S. Kwon, W.B. Kim, et al., Energy consumption reduction technology steel charge considering uncertain metal elements concentrations and production
in manufacturing - A selective review of policies, standards, and research, Int. J. scheduling under time-of-use electricity tariff, J. Cleaner Prod. 176 (2018) 800–812.
Precis. Eng. Man. 10 (5) (2009) 151–173. [30] H. Hadera, I. Harjunkoski, G. Sand, et al., Optimization of steel production schedul-
[2] R. Blundell, R. Griffith, J. Van Reenen, Market share, market value and innovation ing with complex time-sensitive electricity cost, Comput. Chem. Eng. 76 (2015)
in a panel of British manufacturing firms, Rev. Econ. Stud. 66 (3) (2010) 529–554. 117–136.
[3] H. Luo, B. Du, G.Q. Huang, et al., Hybrid flow shop scheduling considering machine [31] Y. Tan, S. Liu, Models and optimisation approaches for scheduling steelmaking–re-
electricity consumption cost, Int. J. Prod. Econ. 146 (2) (2013) 423–439. fining–continuous casting production under variable electricity price, Int. J. Prod.
[4] S. Knopp, S. Dauzère-Pérès, C. Yugma, A Batch-oblivious approach for complex Res. 52 (4) (2014) 1032–1049.
job-shop scheduling problems, Eur. J. Oper. Res. 263 (1) (2017) 50–61. [32] D. Gajic, H. Hadera, L. Onofri, et al., Implementation of an integrated production
[5] V. Fernandezviagas, R. Ruiz, J.M. Framinan, et al., A new vision of approximate and electricity optimization system in melt shop, J. Cleaner Prod. 155 (1) (2016)
methods for the permutation flowshop to minimise makespan: state-of-the-art and 39–46.
computational evaluation, Eur. J. Oper. Res. 257 (3) (2017) 707–721. [33] H. Zhang, F. Zhao, J.W. Sutherland, Scheduling of a single flow shop for minimal en-
[6] L. Merkert, I. Harjunkoski, A. Isaksson, et al., Scheduling and energy – Industrial ergy cost under real-time electricity pricing, J. Manuf. Sci. E. 139 (1) (2017) 014502.
challenges and opportunities, Comput. Chem. Eng. 72 (2015) 183–198. [34] K. Nolde, M. Morari, Electrical load tracking scheduling of a steel plant, Comput.
[7] W. Saad, Z. Han, H.V. Poor, et al., Game-theoretic methods for the smart grid: an Chem. Eng. 34 (11) (2010) 1899–1903.
overview of microgrid systems, demand-side management, and smart grid commu- [35] E. Zitzler, M. Laumanns, L. Thiele, SPEA2: improving the strength pareto evolu-
nications, IEEE Signal Proc. Mag. 29 (5) (2012) 86–105. tionary algorithm for multiobjective optimization, Evolutionary Methods for Design,
[8] J.M. Godoy-Alcantar, J.A. Cruz-Maya, Optimal scheduling and self-generation for Optimization and Control with Applications To Industrial Problems, 2001.
load management in the Mexican power sector, Electr. Pow. Syst. Res. 81 (7) (2011) [36] F. He, K. Shen, L. Guan, et al., Research on energy-saving scheduling of a forging
1357–1362. stock charging furnace based on an improved SPEA2 algorithm, Sustainability 9 (12)
[9] M. Li, D.M. Lei, J.C. Cai, Two-level imperialist competitive algorithm for energy-ef- (2017) 2154.
ficient hybrid flow shop scheduling problem with relative importance of objectives, [37] B. Wang, Z. Guan, S. Ullah, et al., Simultaneous order scheduling and mixed-model
Swarm Evol. Comput. 49 (2019) 34–43. sequencing in assemble-to-order production environment: a multi-objective hybrid
[10] M. Aghelinejad, Y. Ouazene, A. Yalaoui, Production scheduling optimisation with artificial bee colony algorithm, J. Intell. Manuf. 28 (2) (2017) 1–18.
machine state and time-dependent energy costs, Int. J. Prod. Res. 56 (16) (2018) [38] K. Michalak, Evolutionary algorithm with a directional local search for multiob-
5558–5575. jective optimization in combinatorial problems, in: Proceedings of the Genetic and
[11] C. Lu, L. Gao, X.Y. Li, et al., Energy-efficient permutation flow shop scheduling prob- Evolutionary Computation Conference Companion, 2017, pp. 7–8.
lem using a hybrid multi-objective backtracking search algorithm, J. Clean. Prod. [39] K. Deb, H. Jain, An evolutionary many-objective optimization algorithm using ref-
144 (2017) 228–238. erence-point-based nondominated sorting approach, Part I: solving problems with
[12] Y. Liu, H.B. Dong, N. Lohse, et al., A multi-objective genetic algorithm for optimi- box constraints, IEEE Trans. Evolut. Comput. 18 (4) (2014) 577–601.
sation of energy consumption and shop floor production performance, Int. J. Prod. [40] M. Liu, X.N. Yang, F. Chu, et al., Energy-oriented bi-objective optimization for the
Econ. 179 (2016) 259–272. tempered glass scheduling, Omega-Int. J. Manag. Sci. 90 (2020) 21.
[13] S.A. Mansouri, E. Aktas, Minimizing energy consumption and makespan in a two– [41] K. Deb, A. Pratap, S. Agarwal, et al., A fast and elitist multiobjective genetic algo-
machine flowshop scheduling problem, J. Oper. Res. Soc. 67 (11) (2016) 1382–1394. rithm: NSGA-II, IEEE Trans. Evolut. Comput. 6 (2) (2002) 182–197.
[14] J. Cheng, F. Chu, M. Liu, et al., Bi-criteria single-machine batch scheduling with [42] Q. Zhang, H. Li., in: MOEA/D: A Multiobjective Evolutionary Algorithm Based on
machine on/off switching under time-of-use tariffs, Comput. Ind. Eng. 112 (2017) Decomposition, IEEE Press, 2007, pp. 712–731.
721–734.
[15] A. Sharma, F. Zhao, J.W. Sutherland, Econological scheduling of a manufacturing
enterprise operating under a time-of-use electricity tariff, J. Clean. Prod. 108 (2015)
256–270.
17