Energy 125 (2017) 681e704

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Review

Review: Multi-objective optimization methods and application in

energy saving
Yunfei Cui a, b, Zhiqiang Geng a, b, *, Qunxiong Zhu a, b, Yongming Han a, b, **
a
College of Information Science & Technology, Beijing University of Chemical Technology, Beijing, China
b
Engineering Research Center of Intelligent PSE, Ministry of Education of China, Beijing 100029, China

a r t i c l e i n f o a b s t r a c t

Article history: Multi-objective optimization problems are difficult to solve in that the optimized objectives are usually
Received 14 September 2016 conflicting with each other. It is usually hard to find an optimal solution that satisfies all objectives from
Received in revised form the mathematical point of view. Unlike analytical methods and classical numerical methods, which
2 February 2017
require strict mathematical calculation or defined initial search values, intelligent optimization algo-
Accepted 28 February 2017
Available online 2 March 2017
rithms are heuristic algorithms able to find global optimal solutions. In this paper, we make a brief
introduction of multi-objective optimization problems and some state-of-the-art intelligent algorithms.
In order to get the final optimal solution in the real-world multi-objective optimization problems, trade-
Keywords:
Multi-objective optimization
off methods including a priori methods, interactive methods, Pareto-dominated methods and new
Intelligent optimization algorithms dominance methods are utilized. Moreover, we give a review of multi-objective optimization methods
Trade-off solution application in the environmental protection fields, for optimization objectives of energy saving, emis-
Energy saving sions reduction and cost reduction, etc. At last, a whole summary about current difficulties existed in the
Emissions reduction multi-objective optimization problem is given out, serving as suggestions or guidance for future
researches.
© 2017 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
2. Multi-objective optimization problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
2.1. Multi-objective optimization problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
2.2. Multi-objective optimization solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683
3. Multi-objective optimization algorithms and performance test functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683
3.1. Multi-objective optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 683
3.1.1. Biology inspired algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 684
3.1.2. Physics inspired algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686
3.1.3. Geography inspired algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686
3.1.4. Social culture inspired algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686
3.2. Multi-objective optimization test functions and performance evaluation indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
3.2.1. Multi-objective optimization test functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 687
3.2.2. Multi-objective optimization performance evaluation indexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 688
4. Multi-objective optimization trade-off methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
4.1. A priori methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 689
4.1.1. Weighted Sum Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690
4.1.2. ε-Constraint method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690
4.1.3. Objective Programming Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690

* Corresponding author. College of Information Science & Technology, Beijing University of Chemical Technology, Beijing, China.
** Corresponding author. College of Information Science & Technology, Beijing University of Chemical Technology, Beijing, China.
E-mail addresses: [email protected] (Z. Geng), [email protected] (Y. Han).

http://dx.doi.org/10.1016/j.energy.2017.02.174
0360-5442/© 2017 Elsevier Ltd. All rights reserved.
682 Y. Cui et al. / Energy 125 (2017) 681e704

4.2. Interactive methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690

4.3. Pareto-dominated methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 691
4.4. New dominance methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 692
5. Application of multi-objective optimization methods in energy saving and emissions reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 692
5.1. Application in renewable energy resources production optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 692
5.1.1. Application in stand-alone energy systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
5.1.2. Application in hybrid or integrated energy systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 693
5.1.3. Application in distributed generation networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 694
5.2. Application in energy saving optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 695
5.2.1. Application in living activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 695
5.2.2. Application in production processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 695
5.3. Application in emissions reduction optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696
5.3.1. Application in emissions reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 696
5.3.2. Application in ecological design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 697
6. Current difficulties and future directions for researches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 698
7. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 699
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700

1. Introduction Besides, countries around the world have already started to

focus on issues of energy saving and emissions reduction, which is
In the world today, people are paying more and more attention an important way to reach the economic and environmental ob-
to cost minimization, output maximization, energy saving, envi- jectives. As classical energy resources like coal, oil, natural gas, etc.
ronmental protection, and sustainable development issues, etc. in have decreased in amount and renewable resources like wind, solar
various fields. Therefore, us human beings should manage our energy, recyclable water, etc. are under development, how to
production, manufacturing, experiments, and living activities in a improve energy efficiency and effectively utilize limited energy
more efficient and friendly way. Furthermore, these improvement becomes an important consideration in industrial processes and
problems in the scientific research and engineering fields boil daily lives. On the other hand, production and living activities of
down to optimization problems of materials, products, resources human beings have already posed big influences on the earth,
and energy utilization. However, along with the competitive especially bad influences. For the sustainable development of earth
development of technologies and economies around the world, as well as all beings on the earth, environmental protection has
optimization problems become more complicated with larger scale, been and will be of great importance. Consequently, realizing en-
more highly nonlinearity, more objects and constraints. ergy saving and emissions reduction of industrial and living activ-
Optimal problems can be continuous or combinatorial (with ities is beneficial for sustainable development.
continuous or discrete decision variables) optimization problems, Therefore, in this paper, we give an overall systematic overview
constrained or unconstrained optimization problems, linear or about multi-objective optimization methods and application in
nonlinear optimization problems, static or dynamic (off-line or on- energy saving. The paper is organized as follows: Section 2 makes
line) optimization problems, single objective or multi-objective the general definition of the multi-objective optimization problems
optimization problems [1]. To solve optimization problems, we and solutions. Section 3 describes in general developments of
usually take two steps: firstly, we model the optimization problem optimization algorithms, which are able to solve multi-objective
and transfer it into regular expressions, thus to define the decision optimization problems and produce the optimal solutions as a
space, object space and the mapping relation between them; then set. In the following, Section 3 lists theoretical test functions for
we use appropriate and efficient methods to solve those regular evaluating the performance of multi-objective optimization algo-
expressions to get the optimal solution or solutions. rithms. Section 4 introduces common trade-off methods for
Most of the real-world problems, including design, optimization, achieving the final trade-off solution based on preferences of de-
scheduling and control etc. are inherently characterized by multiple cision makers or historical experiences. Besides, the brief applica-
conflicting objectives realization. When addressing these problems, tions of multi-objective optimization methods in various energy-
the parameters or variables are frequently imprecise due to un- consuming industrial processes, for optimization objectives of en-
controllable factors, leading to more complicated problem formu- ergy saving and emissions reduction are presented in Section 5.
lation. Unlike single objective optimization, the most difficulty in Furthermore, Section 6 gives the summary of current difficulties
solving multi-objective optimization problems lies in that there and future directions for research on multi-objective optimization
does not exist a mathematical explainable solution because of the methods and application in energy saving, while Section 7 makes
highly complex nature of simultaneously satisfying several objec- the whole conclusion.
tives, which are usually conflicting to each other. Thus, it is neces-
sary to determine a set of points that represent the optimal solutions 2. Multi-objective optimization problems
and form the Pareto frontier [2]. All points along the frontier are
mathematically equal and valid, and resulting in different values for 2.1. Multi-objective optimization problems
each objective. Although the achieved solutions are of no prefer-
ence, from a practical viewpoint, decision makers usually need only Optimization problems are one of the most common and pri-
one final solution on the basis of their own preferences for the mary problems in both scientific research and engineering prac-
conflicting and incommensurable objectives [3,4]. tice. According to the number of optimized objective function,
 Y. Cui et al. / Energy 125 (2017) 681e704 683

optimization problems can be classified into two categories: single called a global Pareto optimal solution, or optimal solution for
objective optimization problem and multi-objective optimization short. All global Pareto optimal solutions constitute a global Pareto
problem. For multi-objective optimization problems (MOPs), two optimal set, expressed as PS*.
or more objective functions need to be calculated simultaneously. Definition 2.5, Pareto optimal front: The Pareto optimal set
Moreover, these objective functions are always contradictory to presented in the objective function space is called Pareto optimal
each other. A solution good for one function may be bad for another front, which is expressed in the following way:
function or other functions. Therefore, it is hard for a multi-
  
objective problem to search for a solution that satisfies all objec- PF * ¼ f x* x* 2PS*
tive functions. In this situation, there exists a set of feasible
Definition 2.6, non-dominated solution: In the computation
solutions.
process of evolutionary algorithms, the optimal solution in each
The expressions for describing different kind of optimization
generation of evolution population is called non-dominated solu-
objectives may be different, either be maximum functions or be
tion. All non-dominated solutions constitute non-dominated so-
minimum functions. These two extremum functions can be trans-
lutions (abbreviated as NDS). Multi-objective optimization aims at
ferred to each other through the following equation [2].
searching for such non-dominated solutions to approximate real
maxff ðxÞg⇔minf  f ðxÞg (1) optimal solutions.
As shown in Fig. 1., the relationship between the design space
Therefore, any multi-objective optimization problem can be and the objective space was demonstrated, as well as the Pareto
expressed as the following common mathematical model: optimal solutions definition. In the left graph was the simplified
decision space, while the right graph was the objective function
miny ¼ f ðxÞ ¼ ½f1 ðxÞ; f2 ðxÞ; …; fm ðxÞT (2) space. Certain mapping relations were established between these
two spaces. The infeasible regions drawn in both spaces were re-
s:t gi ðxÞ  0ðj ¼ 1; 2; …; pÞ gions out of feasible solutions, which were determined by con-
hk ðxÞ ¼ 0ðk ¼ 1; 2; …; qÞ strains of the optimization problems. The six points listed in the
xmin
i  xi  xmax
i ði ¼ 1; 2; …; nÞ (3) decision space was mapped into the objective function space and
x ¼ ½x1 ; x2 ; …; xn T 2Q the dominance relations were presented as: A_D, A_C, D_C,
y ¼ ½y1 ; y2 ; …; ym T 2j B_F, B_C. The point E was an infeasible solution. A and B points
were non-dominated solutions in the Pareto front, and there were
where in, m is the number of optimized objective functions, Qis a n- dominance relations neither between A and B, nor between D and F
dimensional search space and is determined by the upper bound points.
xmax ¼ ½xmax ; xmax ; …; xmax T and the lower bound xmin ¼ Multi-objective optimization methods try to get solutions that
1 2 n
½xmin ; x min ; …; xmin T of decision variables x ði ¼ 1; 2; …; nÞ. Jis the are as close as possible to the Pareto optimal front and are also
1 2 n i
m-dimensional vector space of objective functions and is deter- uniformly distributed. Such methods will have good convergence
mined byQand the objective function f ðxÞ. Equations gj ðxÞ  and diversity. After the Pareto optimal set is found, decision makers
0ðj ¼ 1; 2; …; pÞ and hk ðxÞ  0ðk ¼ 1; 2; …; qÞ denote p inequality need to select from these optimal solutions the final solution ac-
constraints and q equality constraints. Especially, if p ¼ q ¼ 0, then cording to concrete optimization problems or personal preferences.
the problem is simplified as an unconstrained multi-objective
optimization problem. 3. Multi-objective optimization algorithms and performance
test functions
2.2. Multi-objective optimization solutions
3.1. Multi-objective optimization algorithms
In multi-objective optimization problems, a primary issue is
how to define the solutions. From the viewpoint of theoretical Up till now, there are mainly two kinds of methods to solve
mathematics, there is not one single solution for multi-objective optimization problems [1]: analytical method and numerical
problems, but there are a set of solutions. In 1951, Koopmans [5] method. The analytical method involves strict mathematical proofs
first put forward the concept of Pareto efficient solution set, and derivation, and it can reach exact solution. However, the
which effectively described the solution under the relationship of method is also strict with the problem characteristics, which many
the partial order but not the total order. realistic problems could not match. The numerical method is
Definition 2.1, feasible solution: For a solution vector x2Q, if it designed with appropriate iteration formulas and applied with a
satisfies both inequality constraints and equality constraints for all series of iterations to get the approximate solution. It needs only
j ¼ 1; 2; …; p and k ¼ 1; 2; …; q, it is defined as a feasible solution, defined decision variables and objective variables feedback from
else it is a infeasible solution. optimization problems. The optimization problem can be a black-
All feasible solutions construct a set named feasible domain U. box problem without obvious expressions and it is more suitable
All infeasible solutions construct the set named infeasible domain for realistic problems.
U. Apparently, U þ U ¼ Q, where U4Q and U4Q. Among the numerical methods [6], classical methods like
Definition 2.2, decision variable domination: For two vectors in Newton iteration method, simplex method, conjugate direction
the decision variable space a ¼ ½a1 ; a2 ; …; an T and b ¼ ½b1 ; method, etc. are oriented at single objective optimization problems.
b2 ; …; bn T , if ci2f1; 2; …; mg; fi ðaÞ  fi ðbÞ; and d j2f1; 2; …; mg; These methods have high searching efficiency and fast convergence
fj ðaÞ < fj ðbÞ, then we say that b is dominated by a, expressed as b3a. speed, because they usually start with a given initial point and
Definition 2.3, objective function domination: For two vectors in calculate the next iteration point according to the descending in-
the objective function space g ¼ ½g1 ; g2 ; …; gm T and h ¼ ½h1 ; formation such as gradient. However, the methods have difficulties
h2 ; …; hm T , if ci2f1; 2; …; mg; g i  hi ; and d j2f1; 2; …; mg; in solving problems whose gradient information cannot or need too
g j < hj , then we say that h is dominated by g, expressed as h3g. much to be calculated. When applied for solving multimodal
Definition 2.4, Pareto optimal solution: If the vector objective functions, classical numerical methods are easy to fall into
x ¼ fx1 ; x2 ; …; xn g satisfies the condition :dx2Q; x_x , then x is and hard to escape from local optimum solutions. Moreover, the
684 Y. Cui et al. / Energy 125 (2017) 681e704

Fig. 1. Relationship between the design space and the objective space and solution definition of a two-objective problem.

methods find only one single solution in each iteration step, and the The development process of classical genetic algorithms and
accuracy of the solution depends mostly on the setting of initial differential evolution algorithms are respectively demonstrated in
values. Tables 1 and 2.
Intelligent numerical methods, as one sort of heuristic search Moreover, Jia et al. [31] added chaos neighborhood searching
algorithm, are inspired by behaviors, reactions and communication mechanism to DE to improve its search ability in the early search
mechanisms in nature. Thus developed optimization algorithms are stage and exploration ability in the later search stage. Liu et al. [32]
broadly divided into four categories [7] as shown in Fig. 2: Biology combined DE with PSO to form a hybrid algorithm, which improved
inspired algorithms, Physics inspired algorithms, Geography the performance and accelerated the searching efficiency. The
inspired algorithms, and Social culture inspired algorithms. hybrid algorithm worked especially well for solving constrained
optimization problems.
3.1.1. Biology inspired algorithms Membrane computing [33], also known as a P system, is non-
Biology inspired algorithms are inspired from biological activ- deterministic and distributed parallel computing device, which is
ities in both micro and macro world (such as evolution behaviors), abstracted from the structure and functioning of living cells, as well
or from substantial development and structural features [8]. They as from the interactions of living cells in tissues or neuros. It was
are generally divided into two types: evolution based algorithms initiated by P
aun in 1998, with the first paper published in 2000
and swarm based algorithms [9]. [34]. The structure is consisted of several cell-like membranes,
placed inside a solo skin membrane. Multisets of objects are placed
3.1.1.1. Evolution based algorithms. Evolution based algorithms, in the regions delimited by hierarchical or more general arrange-
also known as Evolutionary Algorithms (EA) are stochastic search ments of membranes, as shown in Fig. 3. The evolution processes of
methods that mimic the survival of the fittest process of natural each object are done in a parallel manner. At last, the evolved result
ecosystems. The algorithms have strong adaptability and self- is output from the skin membrane to the environment. There are
organization, including Evolutionary Programming (EP) [10], mainly three types of P systems: cell-like P systems; tissue-like P
Evolutionary Strategy (ES) [11], Genetic Algorithm (GA), Differential systems and neural-like P systems [35].
Evolution Algorithm (DE), Harmony Search Algorithm (HS), Mem- It has been proved that any Turing computable problems can be
brane Computing (P system), etc. [12]. solved by P systems. The P system has already applications in

Fig. 2. Classification of intelligent optimization algorithms.

 Y. Cui et al. / Energy 125 (2017) 681e704 685

Table 1
Important developments of genetic algorithms.

Genetic algorithms developments Initially proposed time Proposers Run-time complexity [2,26]

GA 1975 Holland [13]

VEGA (vector-evaluated genetic algorithm) 1985 Schaffer [14]
MOGA (Multiobjective Genetic Algorithm) 1993 Fonseca and Fleming [15] OðGmN2 Þ
NSGA (Non-dominated Sorting Genetic Algorithm) 1993 Srinivas and Deb [16] OðGmN2 Þ
NPGA (Niched Pareto Genetic Algorithm) 1994 Horn and Nafpliotis [17] OðGmN2 Þ
SPEA (Strength Pareto Evolutionary Algorithm) 1999 Zitzler and Thiele [18] OðGmðN þ AÞ2 Þ
PAES (Pareto Archived Evolution Strategy) 2000 Knowles and Corne [20] OðGN logM1 A log logAÞ
PESA (Pareto Envelope-Based Selection Algorithm) 2000 Corne, Knowles and Oates [21] OðGN logM1 A log logAÞ
PESA-II 2001 Knowles, Jerram, Corne, et al. [22] OðGmNAÞ
NPGA2 2001 Erichson et al. [23]
Micro-GA 2001 Coello Coello et al. [24]
SPEA2 2002 Zitzler, Laumanns and Thiele [19] OðGmðN þ AÞ2 Þ
NSGA-II 2002 Deb et al. [25] OðGN logM1 NÞ

Table 2
Variants of differential evolution algorithms.

Differential evolution algorithms Initially proposed time Proposer

DE 1995 Storn and Price [27]

FADE (adaptive differential evolution algorithm based on fuzzy theory) 2005 Liu and Lampinen [28]
SaDE (scaling factor adaptive adjustment differential evolution) 2009 Qin et al. [29]
MOSaDE (multi-objective SaDE) 2009 Huang [30]

Fig. 3. A membrane structure.

computer graphics, computer science, cryptography, mathematics, Swarm based algorithms are inspired from social nature and model
abstract chemistry, biology, ecology, artificial intelligence, approx- the collective behavior of populations, such as honey bees, ant
imate optimization, and even linguistics, etc. [36e42]. colonies, and bird flocks, etc. Among these agents (swarm in-
HS was originally developed by Geem [43] for discrete-variable dividuals), they cooperate with each other to search for food,
problems and then expanded to continuous-variable problems. It necessary for their survival, and also keep safe from other agents.
mimics musician's behaviors such as random play, memory-based Swarm based algorithms are consisted of Particle Swarm Optimi-
play and pitch adjusted play to get a perfect state of harmony. zation Algorithm (PSO), Artificial Bee Colony Algorithm (ABC) [45],
Wang et al. [44] proposed a differential harmony search (DHS) al- Artificial Immune System (AIS), Teaching-Learning Based Optimi-
gorithm, combining the mechanisms of differential evolution with zation algorithm (TLBO), Ant Colony Optimization Algorithm (ACO)
harmony search. The DHS enhances the exploration ability of the [46], Cuckoo Search algorithm (CS) [47], Firefly Algorithm (FA) [48],
algorithm. Bacteria Foraging Optimization algorithm (BFO) [49], Coral Reef
Optimization algorithm (CROA) [50], Shuffled Frog Leaping Algo-
3.1.1.2. Swarm based algorithms. In a broadly defined way, swarm rithm (SFLA) [51], Pigeon Inspired Optimization (PIO) [52], etc.
based algorithms are included in evolution based algorithms. In 1987, Christopher initially proposed the concept of Artificial
686 Y. Cui et al. / Energy 125 (2017) 681e704

Life, which meant the system that mimic the behavior character- of physics. The common physics inspired algorithms include
istics in the nature life system, by the way of computers or other Chaotic Optimization Algorithm (COA), Intelligent Water Drops
non-biological media [53,54]. The above mentioned evolution- Algorithm (IWD) [68], Magnetic Optimization Algorithm (MOA)
based algorithms are inspired by Darwinian evolution whereas [69], Gravitational Search Algorithm (GSA) [70,71], Simulated
the swarm intelligence is generated by imitating the behaviors of Annealing (SA) [72,73], etc.
social swarms [55]. Swarm intelligence [56] meant the intelligent
behaviors presented by simple behaviors of individuals in the 3.1.2.1. Chaotic Optimization Algorithm (COA). The random search
population without central control. These individuals behaved to technique was introduced by Hamzacebi and Kutay [74], which was
solve the foraging, searching and visiting, transportation and adaptable to different optimization problems and the simplest
transmission problems. Based on the swarm intelligence, swarm- heuristic algorithms. As introduced by Vela'squez Henao [75], the
based algorithms were then produced. use of chaotic sequences instead of quasi-random numbers seemed
3.1.1.2.1. Particle swarm optimization (PSO). In 1995, Kennedy to be a more powerful strategy for improving many traditional
and Eberhart [57,58] proposed the PSO algorithm, whose central heuristic algorithms, because the chaotic sequences have charac-
idea was information sharing mechanism. The PSO has been widely teristics of ergodicity, randomness, and regularity. An essential
used in various fields to solve different kinds of optimization feature of chaotic systems is that small changes in the parameters
problems. Many variants of the PSO algorithm have been proposed or the initial values lead to vastly different future behaviors [76].
to maintain or strengthen the diversity, so as to escape from local The chaos optimization algorithm (COA) was first proposed in
optima or premature convergence. Li et al. [59] combined PSO with 1997 by Li et al. [77], in which the Logistic map was introduced to
NSGA-II and the experiment results showed that the combined produce chaos variables as optimization variables. Besides Logistic
method had better performance than NSGA-II. Coello Coello et al. map, other mapping methods were also incorporated in the COA,
[60] proposed the multi-objective PSO (MOPSO), which incorpo- such as Tent map [78,79]. The COA was also combined with other
rated external population with adaptive grids. algorithms to form hybrid chaos optimization methods [80e82].
3.1.1.2.2. Teaching-learning based optimization (TLBO). The TLBO
algorithm was first proposed by Rao et al. [61]. There are two 3.1.3. Geography inspired algorithms
phases in the TLBO: teacher phase and learner phase. Learners learn Geography inspired algorithms are one sort of metaheuristic
from the teacher in the teacher phase and from each other in the algorithm and generate random solutions in the geographical
learner phase. The teacher is considered as the best solution in the search space. These optimization algorithms are classified as Tabu
entire population obtained thus far. In order to get the global Search Algorithm (TS), Imperialistic Competition Algorithm (ICA),
optimal solutions, a modified TLBO algorithm was proposed by etc.
Hosseinpour et al. [62], which added a mutation process similar to
that of DE. Niknam et al. [63] proposed a modified TLBO algorithm 3.1.3.1. Imperialistic Competition Algorithm (ICA). The ICA is
with two mutation operations and two crossover operations added, inspired by imperialism, which is the policy of extending power
so as to enhance the local and global search abilities of the algo- and signifies the role of a government [83]. The number of colonies
rithm. TLBO also shows good performance in solving large scale determines the power of an imperialist. Strengthening the au-
optimization problems with little computational efforts [64]. thority of an imperialist makes other imperialists weaker.
3.1.1.2.3. Artificial Immune System (AIS). The AIS is inspired from
immunology and acts as an adaptive system that mimics the 3.1.3.2. Tabu Search Algorithm (TS). Tabu search algorithm was first
function, principles and model of immunology to solve complicated suggested by Glover [84] in 1986, which was a meta-heuristic
problems [65]. It has been successfully applied in fields of anomaly search method based on local search. It explores all feasible solu-
detection, computer security, data mining, optimization, etc. In tions in the search space by a sequence of moves. Especially, a set of
Table 3 illustrated are developments of AIS. Wherein, the NNIA is moves are forbidden at each iteration step to escape from local
especially advantageous in solving many-objective optimization minima [85].
problems with more than three optimization objectives.
Compared to classical numerical methods, broadly defined 3.1.4. Social culture inspired algorithms
evolution based algorithms is one sort of probability search algo- Social culture inspired algorithms are inspired by the social,
rithm based on population. These algorithms need neither extra economic and cultural systems etc. that incorporate the cultural
initial points nor gradient information of objective functions. evolution theory into optimization algorithms.
Therefore, they are suitable for optimization problems that cannot As shown in Table 4, there are some classical developments of
be solved by classical numerical methods. Moreover, evolution social culture inspired algorithms.
based algorithms have characteristics of parallelism and distribu- In 1989, Moscato firstly proposed the Memetic Algorithm, which
tion, appropriate for solving large-scale/high-dimensional optimi- used the local heuristic search to imitate the mutation process
zation problems. backed up by large amount of professional knowledge. The Gran-
ular Computing mimics the human thoughts from different levels
3.1.2. Physics inspired algorithms of granules. It is based on the space partition of problem concepts,
Physics inspired algorithms for optimization problems are also able to effectively analyze and deal with problems of fuzziness,
heuristic algorithms. They imitate the physical behaviors and non-accuracy, non-consistency and partial true values.
properties of the matters or follow the same philosophy as the laws In summary, Fig. 4 demonstrates the overall development

Table 3
Developments of the artificial immune system.

Developments of the artificial immune system Initially proposed time Proposer

AIS 1998 Dasgupta [65]

MISA (multi-objective immune system algorithm) 2005 Coello Coello et al. [66]
NNIA (non-dominated neighborhood immune algorithm) 2008 Gong et al. [67]
 Y. Cui et al. / Energy 125 (2017) 681e704 687

Table 4
Developments of social culture inspired algorithms.

Social culture inspired algorithms Source of inspiration Initially proposed time Proposer

MeA (Memetic Algorithm) Social culture 1989 Moscato [86]

CuA (Cultural Algorithm) Social culture (signal, knowledge, etc.) 1994 Reynolds [87]
SCO (Social Cognitive Theory) Social cognitive process 2002 Xie [88]
SGA (Selfish Gene Algorithm) Selfish gene in human beings 1998 Corno et al. [89]
GrC (Granular Computing) Information cognition 1996 Lin [90]
AC (Affective Computing) Social and cultural affection, etc. 1997 Picard [91]

problems considering energy, economics and environment issues in

processes that produced or consumed energy [7]. As renewable
energy resources are clean and environment-friendly, they are paid
enormous attention in recent years. PSO [98] or its improved vari-
ants like MOPSO [99], AMOPSO [100], BB-MOPSO [101], and LAPSO
[102] etc. were tested efficient in energy optimization applications.
Meanwhile, GA, with most accepted and applied variants
(including VEGA, MOGA, SPEA, SPEA2, NSGA, NSGA-II, PESA, PESA-
II, NPGA, NPGA2, etc.) were also used or improved varying with
concrete multi-objective optimization problems, like building en-
ergy optimization [103,104], distribution transformer optimal
design [105], complex industrial processes [106e108], or processes
integrated renewable resources [109].
New methods are proposed every day to solve multi-objective
optimization problems. Besides, combined methods based on the
classical optimization method and the intelligent optimization al-
gorithm or on two kinds of intelligent algorithms are being
developed for various real-world multi-objective optimization
problems. Intelligent algorithms applications in energy and envi-
ronment protection optimization issues will be discussed in detail
in Section 5.

3.2. Multi-objective optimization test functions and performance

evaluation indexes

3.2.1. Multi-objective optimization test functions

As it is hard to evaluate the performance parameters of intelli-
gent algorithms theoretically, researchers generally use test func-
tions to verify the algorithm performances. Zitzler et al. [110]
constructed a set of test problems called ZDT test function set,
Fig. 4. Development process of common intelligent optimization algorithms. which was consisted of six problems ZDT1~ZDT6. These problems
are two-objective optimization problems with different forms of
expression and properties. Because their Pareto fronts are known,
process of common intelligent optimization algorithms. At the they are one of the most common used test problems. Therein, ZDT1
same time, description of advantages and disadvantages of some and ZDT4 are convex functions while ZDT2 and ZDT6 are concave
classical intelligent optimization algorithms are listed in Table 5. functions, ZDT3 is a non-continuous function, ZDT4 is a multi-modal
Zhou et al. [92] surveyed the development of MOEAs (multi- function, and ZDT5 is a function with deceptive property.
objective EAs) in detail. In the paper covered included algorithmic Deb et al. [111] constructed a set called DTLZ test function set,
frameworks and applications such as MOEAs with specific search which allowed the decision variables and objective functions to
methods, MOEAs for multimodal problems, constraint handling in extend to any dimension. The DTLZ test function set includes seven
or with MOEAs, computationally expensive multi-objective opti- unconstrained optimization problems DTLZ1~DTLZ7 and two con-
mization problems (MOPs), dynamic MOPs, combinatorial and strained optimization problems DTLZ8~DTLZ9. They are also widely
discrete MOPs, and etc. There was also a summary of the major used for testing the performance of optimization algorithms. Deb
applications of MOEAs in solving real-world problems. et al. [112] also constructed a set of constrained multi-objective
Rodrigues et al. [93], Cambero et al. [94], Chaouachi [95], H.A. optimization problems (called DEB) with different Pareto optimal
et al. [96], and Fadaee et al. [97] reviewed the application of boundaries. Huband et al. [113] defined a set of WFG test problems,
intelligent algorithms (especially EAs) in the multi-objective opti- and constructed a scalable toolkit of test problems. Li and Zhang
mization of economic, energy, environment, or technical issues in [114] proposed a set of continuous test problems whose variables
the fields of wind farm, forest biomass supply chains, micro-grid, were correlative and the Pareto front surface was with arbitrary
distribution generation systems, and hybrid renewable energy complexity, which was able to reflect the complexity in the real-
systems. They found out that intelligent algorithms were utilized world multi-objective optimization problems. There are also
effectively able to find global optima. some other common test problems like Schaffer's study (SCH) [115],
The PSO algorithm was one of the most popular intelligent al- Fonseca and Fleming's study (FON) [116], Kursawe's study (KUR)
gorithms that were applied for solving multi-objective optimization [117], etc.
688 Y. Cui et al. / Energy 125 (2017) 681e704

Table 5
(Dis)advantages of some classical intelligent optimization algorithms.

Intelligent optimization Advantages Disadvantages

algorithms

GA(NSGA-II) Good diversity of solutions (using the environment selection Premature or long time convergence;
strategy based on neighborhood rules (in SPEA2), or the elitist Probable loss of best solutions;
keeping mechanism); Not guaranteed to find a global optimum.
Uniform distribution of Pareto solutions with good performance
(based on the crowding distance method);
Relatively low computation complexity (fast non-dominated
solution ranking method).
DE Easy to understand and strong robustness; Its parameters need to be preset; Parameters have important
Few parameters need to be adjusted; influences on the algorithm performance.
Able to tackle non-differentiable, non-linear and multimodal
functions promptly.
P System Understandable, scalable and programmable; A relatively new and developing method, not mature in application.
Parallel execution manner;
Adequate for handling discrete processes.
HS Only a few parameters and easy to be implemented. Slow convergence speed and weak search guidance.
PSO Simple procedures and not complex adjustment; Not sure to be strictly convergent;
Easy to be implemented with relatively fast speed; Relatively weak local search ability;
Able to escape from the local optima and find the global optimal Probable to be immersed in local optima in multi-modal problems.
solutions in most cases.
TLBO Good learning abilities and parameter-free. Relatively low local search ability; Premature convergence due to
Good performance in solving large scale optimization problems lack of sufficient information sharing;
with little computational efforts. Easy to be immersed in local optima when dealing with low
dimension and complicated solution space.
AIS Distributed, self-organizing and self-adaptable structure; Parallel Not profound or mature research and application, especially in the
and robust in execution; optimization field.
High search efficiency;
Able to escape from premature convergence;
Suitable for dealing with problems with constraints and multiple
criteria.
GSA Global search, easy execution and implicit parallelism. Need to be improved in search efficiency and solution precision.
SA Simple structure, parallel computing, high efficiency, extensive Slow convergence;
applications and flexible to use; time-consuming in computation; Sensitive to parameters;
Can converge asymptotically to the global optimal solution; May be constrained by initial conditions.
Especially suitable for solving large-scale combinatorial
optimization problems.
COA Simple structure, easy implementation; Dependent on initial values;
Short execution time, high execution efficiency; May not be able or need too much time to find or approximate the
Excellent stochastic searching capability for global optima and optimal solution under unsuitable initial values.
robust mechanisms of escaping from local optima.
ICA High convergence accuracy and fast convergence speed; Relatively low solution precision;
Appropriate for optimization of nonlinear optimization problems Easy to be premature convergent and immersed in local optima.
with high dimensions.
TS A global optimization algorithm; Dependent on initial solutions;
Able to escape from local optima; Serial but not parallel computation process.
Effective in solving combinatorial optimization problems.
MeA Combined priorities of global and local search; The design of local search module has important influence on
High search efficiency, fast convergence speed, parallel working solutions.
manner;
Relatively good diversity and fault tolerance ability.

3.2.2. Multi-objective optimization performance evaluation indexes (1) Convergence Metric: Suppose that the real Pareto optimal
The solutions found by the multi-objective optimization algo- solution set distributed uniformly along the ideal Pareto
rithms are a set of approximate Pareto optimal solutions and we front is P ¼ ðp1 ; p2 ; …; pr Þ, and the approximate Pareto
need to evaluate this set of approximate solutions. The evaluation optimal solution set searched by the intelligent optimization
indexes usually involve the following three indexes: algorithms is A ¼ ða1 ; a2 ; …; as Þ. The calculation expressions
of the Convergence Metric are as follows:
(1) Convergence: the solutions that are most approximate to the
Pareto optimal solutions are the best. X
s
IC b di =s (4)
(2) Uniformity: the good solutions should be distributed uni-
i¼1
formly along the Pareto optimal frontier.
(3) Distribution: the final solutions should cover the whole vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 12ffi
u 0
Pareto optimal frontier as much as possible. u m f a   f p
uX k j k j
di ¼ min t @
max min
A (5)
Deb et al. [118] proposed the Convergence Metric index (IC) to 1jr
k¼1 fk  fk
evaluate the convergence performance of the multi-objective
optimization algorithms. Schott [119] proposed the Spacing wherein, s is the number of searched solutions, r is the number of
Metric index (IS). real Pareto solutions, m is the number of objective functions, di is
 Y. Cui et al. / Energy 125 (2017) 681e704 689

the minimal normalization Euclidean Distance between each non-commensurate objectives, more and more researches have
searched solution ai and the real Pareto optimal solution set, been done on from classical optimization algorithms to intelligent
fkmax and fkmin are maximal and minimal values in the set P of the kth optimization algorithms.
objective. In the past, the earliest and direct method for dealing with
The smaller the value IC is, the better convergence the algorithm multi-objective problems was to transfer them into single objective
has and the closer to the real Pareto front. problems, and then used classical optimization algorithms to solve
the problems. In applying this method, we need decide the
(2) Spacing Metric: importance degree of each objective, which is previously deter-
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi mined using a priori method (like Weighted Sum Method) or
u s  2
u 1 X determined during the search process using the interactive method
IS bt d  di (6) (like Boundary Intersection Method) [6]. As shown in Fig. 5, there is
s  1 i¼1
a summarized demonstration of trade-off methods for solving
multi-objective optimization problems.
( )
X
m   
di ¼ min fk ðai Þ  fk aj  ; ai ; aj 2A; i; j ¼ 1; 2; …; s (7)
1js
k¼1 4.1. A priori methods
If IS ¼ 0, it denotes that the searched non-dominated solutions
in the objective space is distributed evenly spaced. There several kinds of a priori methods such as Weighted Sum
Moreover, there are also other evaluation indexes like Error ratio Method, Constraint Method, Objective Programming Method, Dic-
(IER) [120], Coverage of two sets (ICS) [121], Generational Distance tionary Ordering Method, Analytic Hierarchy Method and etc.
(IGD) [122], Hypervolume evaluation index (IH) [123], ε evaluation
index (Iε) [123], etc.

(3) Error ratio: Error ratio is used to check how many approxi-
mate Pareto optimal solutions are covered by real Pareto
optimal solutions. The ratio of the number of solutions that
are not covered to the whole population is called Error ratio
IER. Thus, if IER ¼ 0, it means that all approximate Pareto
optimal solutions are in the real solutions; IER ¼ 1 denotes
that all approximate solutions are out of the real solutions.
(4) Coverage of two sets: This index is used to evaluate the
mutual coverage rates of two solution sets (A and B) and
transfer the coverage rates into a pair of real numbers in
[0e1].

Ics ðA; BÞ ¼ jfb2Bjda2A : a7bgj=jBj (8)

Ics ðB; AÞ ¼ jfa2Ajdb2B : b7agj=jAj (9)

wherein, ICS(A,B) denotes the ratio of solutions in the set B that are
dominated by or equal to solutions in the set A. ICS(A,B) ¼ 1 means
that B is totally covered by A, ICS(A,B) ¼ 0 means that B is not a little
covered by A. 0<ICS(A,B) < 1 denotes that B is partly covered by A. If
ICS(A,B) ¼ 1 and ICS(B,A) ¼ 0, then we can say that A is better than B.

(5) Generational Distance: This index describes the distance

between the searched Pareto front and the real Pareto front.
vffiffiffiffiffiffiffiffiffiffiffiffiffi,
u s
uX 2
IGD ¼t di s (10)
i¼1

wherein, s is the number of searched solutions, di is the Euclidean

Distance between the solution and the corresponding real Pareto
optimal solution in the objective space. IGD ¼ 0 means that the
approximate solutions are completely close to the real solutions.

(6) Hypervolume evaluation index: This index is defined as the

volume of space that is dominated by the searched solution
set but not by the reference point set. It takes into consid-
eration both convergence and diversity.

4. Multi-objective optimization trade-off methods

In order to get a trade-off solution for multiple, conflicting, and Fig. 5. Classification of multi-objective trade-off optimization methods.
690 Y. Cui et al. / Energy 125 (2017) 681e704

4.1.1. Weighted Sum Method guarantees the objective fk ðxÞ get the ideal value and gives certain
For m optimization objectives fi ðxÞ, i ¼ ð1; 2; …; mÞ, we use consideration the other m-1 objectives fi ðxÞ. However, there exist
Weighted Sum Method to transfer multiple objectives into one difficulties in setting upper bounds of estimated objective function
single objective: values. If the upper bounds are too small, the algorithm may not
find the feasible solution; or if the upper bounds are too large, the
X
m
other objectives as constraints may have excessive losses. Thus, the
f ðxÞ ¼ min ½wi  fi ðxÞ (11)
reasonable values of the upper bounds are generally determined
i¼1
according to engineering experiences.
wherein, wi  0; i ¼ ð1; 2; …; mÞ is a set of weighted coefficients,
Pm
4.1.3. Objective Programming Method
i¼1 wi ¼ 1. Therefore, the multi-objective problem is transferred
into a single objective problem, which can be solved by classical In the Objective Programming Method, decision makers firstly
optimization algorithms. set the desired values of each objective do ¼ ½do1 ; do2 ; …; dom T and
The structure of the Weighted Sum Method is so simple that it is add them into the optimization objectives.
easy to understand and apply. However, in this method, decision  
Pm  
makers need to decide the weighted coefficients beforehand ac-  
min fi ðxÞ  doi 
cording to real-world problems or technical experiences. minf ðxÞ0 i¼1   (14)
A variant of the Weighted Sum Method is the Weighted Product s:t: x2Q
Method, which is represented as follows:
The desired values shall be set within the feasible region so as to
Y
m
wi
reach the Pareto optimal solution efficiently. Similarly, it is also
f ðxÞ ¼ min ½fi ðxÞ (12) difficult to set the desired values. Moreover, the Objective Pro-
i¼1 gramming Method is effective in solving linear programming
Xia and Cui et al. [124] proposed a hybrid chaos-PSO algorithm problems but less effective in solving nonlinear complex problems.
based on sequential quadratic programming (SQP) method (SQP- Even if these classical methods for dealing with multi-objective
CPSO), to realize the optimal modeling of the ethylene cracking optimization problems are simple to use, they can only solve
furnace. The optimized objectives were transformed into one single convex optimization problems through coupling multiple objec-
objective using the WSM according to previous experiences. SQP- tives based on a prior experience, therefore, these methods are
CPSO was used to optimize the single objective. The proposed limited for application [131]. Since 1989 when Goldberg [132]
method improved the accuracy of the Kumar model with certain firstly introduced Pareto theory in economics into the genetic al-
extendable performance. gorithm, more and more intelligent algorithms and also new
In order to maximize methanol production and minimize CO2 dominance mechanisms are proposed and applied to solve multi-
emission for a green integrated methanol case (GIMC), Taghdisian objective optimization problems.
et al. [125] proposed an ecological design method for the sustain-
able development of this methanol production process. The mul- 4.2. Interactive methods
tiple objectives were transferred into one single objective by the
weighted product method (WPM), and GA was applied to get the In order to actively exploit the decision makers' knowledge and
optimal solution. experiences, interactive methods were developed, such as the
Ju et al. [126] made performance analyses for a hybrid energy interactive weighted Tchebycheff method [133], the Light Beam
system from aspects of energy, economic, and environment. They Search [134] and the NIMBUS method [135e137].
used the entropy weight method [127] to calculate the weight co- The interactive methods incorporate preferences of decision
efficients of the objectives so that the multiple objectives were makers for each objective during the optimization process. As
weighted into a single objective. Then the single objective was usual, interactive methods implement an achievement scalariza-
optimized by the GAMS software. tion function (ASF) [138] to generate Pareto optimal alternatives.
There are two widely-applied interactive methods for dealing
4.1.2. ε-Constraint method with multi-objective decision problems: Normal Boundary Inter-
The ε-Constraint Method was proposed by Haimes et al. [128], section (NBI) and Normalized Normal Constraint (NNC) methods.
which required bounds for each objective function to be decided by They are relatively new scalarization methods compared with the
experiences. WSM, which reformulate the multi-objective optimization problem
When applying Constraint Method for dealing with objectives into a parametric single objective optimization problem.
fi ðxÞ, i ¼ ð1; 2; …; mÞ, we select from fi ðxÞ one objective as the In 1998, Das and Dennis [139] proposed the NBI method, which
optimized objective and the others as constraints of the optimiza- tackled the multi-objective problems from a geometrically intuitive
tion problem. viewpoint. The method first builds the convex hull of individual
minima (CHIM) and then constructs (quasi-)normal lines to the
minfk ðxÞ plane. The rationale lies in that the intersection between the (quasi-
minf ðxÞ0 (13)
s:t:fi ðxÞ  εi ; 1  i  m; isk )normal from any point on the CHIM, and the boundary of the
feasible objective space closest to the origin is expected to be the
wherein ε ¼ ½ε1 ; ε2 ; …; εm T is the set of upper bounds of esti- Pareto optimal.
mated objective function values determined by decision makers in The NBI method is able to form a near-uniform spread of the
advance. Pareto-optimal frontier, making the NBI a more attractive approach
Geng and Cui et al. [129,130] applied COA combining with the to the Weighted Sum Method in solving non-convex, high-
constraint method to realize the modeling and optimization of dimensional multi-objective problems.
ethylene cracking furnace, which not only improved the accuracy of Ganesan et al. [140] used the NBI interactive method to
the model with relatively high computation efficiency, but also compromise the multiple optimized objectives in the synthesis gas
improved the energy utilization efficiency of the process. production process of combined carbon dioxide reforming and
The advantage of the Constraint Method lies in that it partial-oxidation of methane technologies. In conjunction with the
 Y. Cui et al. / Energy 125 (2017) 681e704 691

NBI method, the GSA and the PSO algorithms were adopted to Pareto set with predefined accuracy was calculated by the sand-
realize the process optimization of objectives of methane conver- wich approximation method, which can handle non-convexities.
sion, carbon monoxide selectivity and the hydrogen to carbon Then the decision makers navigated interactively on the Pareto
monoxide ratio. The optimization results of these two algorithms set and explored different optimal solutions by the CHEMASIM tool.
were compared using the Euclidean distance metric. The PSO al-
gorithm outperformed the GSA method in terms of uniformity of 4.3. Pareto-dominated methods
the Pareto front and computational efficiency.
In 2003, Messac et al. [141] proposed the NNC method, which Since 1994 [14], when the NSGA was first used to achieve the
was similar as the NBI method but combined with features of the Pareto front for the multi-objective optimization problem, intelli-
ε-constraint method. The ε-constraint method minimizes the most gent optimization algorithms based on Pareto-dominated methods
important objective function fk , while the other objectives are came up one after another. Intelligent algorithms were able to solve
added as inequality constraints with the form fi  εi . Based on this the non-convex optimization problems without coupling objec-
idea, in the NNC method, a plane called the utopia hyperplane is tives, so as to maintain individual features of each objective. More
constructed through all normalized individual minima, and equally and more researches have been done on intelligent optimization
distributed points f p in this hyperplane are determined by algorithms searching for Pareto-dominated solutions for multi-
consistently varying the weights. Then m-1 hyperplanes are con- objective optimization problems.
structed for other objective functions. These hyperplanes are cho- Combined with the newly emerging chemical reaction optimi-
sen perpendicular to each of the m-1 utopia plane vectors, which zation algorithm (CRO) [150], the PSO algorithm created new
join the individual minimum corresponding to the selected objec- molecules (particles) both used by CRO operations and mechanisms
tive fk . Furthermore, there is an Enhanced Normalized Normal of PSO. The crowding distance method was used to keep the di-
Constraint method (ENNC) proposed in Ref. [142]. versity of the Pareto set, which was stored in the external archive.
Logist et al. [143] incorporated the NBI and NNC methods in a The hybrid optimization method with balancing operators was able
deterministic multiple shooting optimal control to mitigate the to avoid premature convergence and explore more in the search
drawbacks of the Weighted Sum Method. The combined interactive space. (CRO is an optimization technique inspired by the chemical
method was able to obtain an equal distribution along the non- reaction process [151]. It imitates the interactions of molecules in a
convex Pareto front. It can deal with equality/inequality con- chemical reaction to reach a low energy stable state.)
straints and boundary value problems, with tight tolerances for Although intelligent optimization algorithms based on Pareto-
global and local optimality. The resulting optimization method is dominated methods have better performance in convergence, di-
successfully used in the design of a chemical reactor and the control versity, robustness and flexibility than classical methods, their
of a bioreactor. computation process is sometimes time-consumed and relatively
The integration of optimization techniques with these interac- low-efficient [152,153].
tive methods were efficiently used to tackle non-convex optimal Xiang and Zhou [154] proposed a dynamic artificial bee multi-
control problems [144] and applied to different engineering fields colony algorithm for multi-objective optimization problems, mak-
[145]. ing use of a multi-deme model and dynamic information exchange
However, NBI and (E)NNC may overlook the extreme parts of the strategy. In order to accelerate the computation process, the
Pareto set. Therefore, Vallerio et al. [146] introduced an Interactive searched non-dominated solutions were stored in an external
Geometric Extension (IGE) technique to extend the Pareto set for archive, and the diversity over the archived solutions was main-
NBI and (E)NNC methods based on geometric considerations. Then tained by using the crowding-distance strategy.
the extended NBI or (E)NNC methods were applied successfully to Chen et al. [155] proposed an adaptive grid particle swarm
three scalar multi-objective problems and the multi-objective optimization (AGPSO) algorithm to solve the PID control parame-
optimal control of a tubular and a fed-batch reactor. The results ters optimization problem of a hydraulic turbine regulating system
demonstrated the low computational burden and applicability to (HTRS) under load or unload conditions. Moreover, the fuzzy
higher than three dimensional problems of the proposed methods. membership function was used to select the optimal trade-off so-
Moreover, Vallerio et al. [147] presented an interactive frame- lution from the Pareto set. The simulation results proved the effi-
work based on NBI and ENNC to realize the nonlinear dynamic ciency and solution quality of the proposed algorithm.
multi-objective optimization. By the active use of Pareto Browser In the optimization procedures, the loss of non-dominated so-
Graphical User Interface (GUI), decision makers expressed their lutions and the appearance of false non-dominated solutions may
preferences via the browsing of scalarization parameters such as disturb the solution search process, thus affect the solution accu-
weights. The parameters were adapted interactively. Finally, the racy and algorithm performance.
introduced interactive framework for multi-objective dynamic Based on the analysis of these problems, a dynamic multi-
optimization was successfully tested for a three and five-objective objective optimization algorithm (DMOOP) inspired by mem-
fed-batch reactor case study with uncertain feed temperature and brane computing was proposed [156]. DMOOP was applied in the
heat transfer parameters. optimal control of a time-varying unstable plant. The key in solving
Most interactive methods for multi-objective optimization dynamic multi-objective optimization problems lied in construct-
problems may impair at least one objective function to get a solu- ing discrete, equivalent, and static sub-problems during each time
tion. Hence, Miettinen et al. [148] proposed a NAUTILUS method range. The experiment results showed that the proposed method
based on the assumptions that past experiences affected decision was able to solve optimal control problems effectively, with short
makers' hopes and decision makers did not react symmetrically to rise time, small overshoot and expected output.
losses and gains. The ability of NAUTILUS to obtain a non-anchored Yu et al. [157] proposed a self-adaptive multi-objective
Pareto optimal solution made it an ideal tool to find an initial so- teaching-learning-based optimization method (TLBO) called SA-
lution for any other interactive schemes. MTLBO. In this method, learners adaptively chose the learning
In order to deal with the objectives impairment problem, Bortz model according to their own knowledge level, performed corre-
et al. [149] integrated an algorithm based on a state-of-the-art sponding search functions to improve the performance of SA-
steady-state flow sheet simulator for designing a distillation pro- MTLBO. Finally, the proposed SA-MTLBO was applied in the
cess for the separation of an azeotropic mixture. Firstly, a minimal naphtha pyrolysis process to realize the maximization of ethylene,
692 Y. Cui et al. / Energy 125 (2017) 681e704

propylene, and butadiene yields, thus improving the energy utili- the multi-objective problem into the single objective problem. The
zation efficiency of the feed oil (i.e. naphtha). proposed methods were helpful for decision making problems with
Niknama et al. [158] presented a multiobjective modified honey uncertainty and hesitation, and were able to solve multi-objective
bee mating optimization (MHBMO) algorithm to realize the eco- optimization problems in production, planning, scheduling and
nomic and energy optimization of the distribution network. They manufacturing systems.
adopted a fuzzy-based decision maker, according to experiences Gharavi et al. [170] utilized the fuzzy logic method in the non-
and preferences, to select the ‘best’ trade-off solution among the autonomous hybrid green power system (HGPS), to compromise
non-dominated optimal solutions. objectives of economics and environmental emission, with
Zheng et al. [159] proposed a multi-objective group search different degrees of importance.
optimizer method for the power dispatch optimization of a large- Zhang and Li [171] combined evolutionary algorithms with
scale integrated energy system (LSIES). Meanwhile, a decision classical multi-objective dealing methods and proposed a multi-
making method based on evidential reasoning (ER) [160] was used objective evolutionary algorithm based on decomposition (MOEA/
to determine the trade-off optimal solution from the Pareto- D) method. The MOEA/D method decomposed the multi-objective
optimal solutions. problems that approximate the Pareto front into a number of single
Wei et al. [161] proposed a multi-objective interval optimization objective sub-problems, and solved these sub-problems using the
model to solve the optimization problem of small-scale integrated evolutionary algorithm. The neighborhood relations between sub-
energy systems. ER was also used in the decision making stage to problems were defined as the distances between weighted vec-
select the final optimal solution. tors of sub-problems. The optimization process of each sub-
Shirazi et al. [162] employed MOGA to solve the simultaneous problem was finished by the cooperation between it and its
optimization problem of thermo-dynamic (energetic and exergetic) neighboring sub-problem.
and economic objectives of an ice thermal energy storage (ITES) Mlakar et al. [172] proposed a surrogate-model-based evolu-
system. The final optimal solution was determined using TOPSIS tionary algorithm, called differential evolution multi-objective al-
(Technique for Order Preference by Similarity to an Ideal Solution) gorithm based on Gaussian process models (GP-DEMO). The new
decision making method to reach a reasonable trade-off between defined relations with bounding box, instead of Pareto domination,
objectives. were used to compare solutions under uncertainty. It was able to
reduce the error ratio due to the inaccurate surrogate model
approximation. By adding the evolutionary algorithm to the
4.4. New dominance methods
surrogate-model, optimal solutions were found, evaluated and
used for updating the model.
Because the Pareto-dominated methods are not able to simul-
Some examples of new dominance methods are listed in Table 6
taneously solve the convergence and distribution problems of
in summary.
optimization algorithms, Laumanns et al. [163] proposed the
ε-dominance mechanism. The ε-dominance mechanism is different
from Pareto-dominance mechanism in that the former one is based 5. Application of multi-objective optimization methods in
on the idea of hyper grids in space. Afterwards, Hern'andez-Dıaz energy saving and emissions reduction
et al. [164] proposed the combined Pareto-adaptive ε-dominance
mechanism. In the mechanism, the ε was a vector whose length was As more and more attention is paid to environmental protec-
the number of the hyper grids. The values of elements in the vector tion and sustainable development in real-world production and
were related to the geometrical shape of the Pareto front. The living activities, pollutants emissions, energy efficiency and eco-
values were adjusted self-adaptively according to domination nomic profits become more and more important considerations
strength information of different parts of the Pareto front. that should not be ignored in both pure energy-consuming fields
Later on, Brockoff et al. [165] researched partial dominance. Deb and energy production fields (especially of renewable and sus-
and Saxena [166,167] combined primary component analysis (PCA) tainable energy resources). These considerations start to be
method, correntropy PCA method with evolutionary algorithms to simultaneously included as multiple optimization objectives in
solve large-dimensional multi-objective problems. In 2007, Koduru real-world problems, but not simply summed up as one single
et al. [168] proposed the concept of fuzzy dominance, which used optimization objective.
fuzzy dominance function to give weights to each objective func-
tion and summed up the weighted objectives. 5.1. Application in renewable energy resources production
To deal with uncertain parameters in real-world problems, optimization
Singh and Yadav [169] proposed a linear ranking function (the ac-
curacy function) to convert the intuitionistic fuzzy model into a Unlike other production processes that are aimed at energy
crisp multi-objective nonlinear programming problem. They used a utilization optimization, renewable energy resources (RESs)
nonlinear membership function with Zimmerman's fuzzy tech- (wind, solar, etc.) production processes are more focused on en-
nique, Gamma-operator and Min-bounded sum operator to transfer ergy production efficiency, mostly represented by the design

Table 6
Examples of new dominance methods.

New dominance methods Initially proposed time Proposer

ε-dominance mechanism 2002 Laumanns, Thiele, Deb, and Zitzler

PCA method 2005 Deb and Saxena
Partial dominance 2006 Brockoff, and Zitzler
Correntropy PCA method 2007 Saxena and Deb
Fuzzy dominance 2007 Koduru, Das, andWelch
MOEA/D method 2007 Zhang and Li
Pareto-adaptive ε-dominance 2007 Hern'andez-Dıaz, Santana-Quintero, Coello Coello, and Molina
 Y. Cui et al. / Energy 125 (2017) 681e704 693

optimization of renewable energy systems. Together with eco- result was compared with that from the HOMER software tool.
nomic or environmental objective, the energy efficiency objective The PSO algorithm used for multi-objective optimization of the
gets more and more attention in various renewable resources system achieved better reduction of the NPC and CO2 emissions,
production processes. These processes include stand-alone en- while realizing effective system size design, and meeting the load
ergy systems like the wind farm, the power grid or micro-grid, demand of desalinated water and electricity. As the scale of the
and the biomass production processes, hybrid energy systems or system became large, the advantage of the PSO algorithm over
distributed generation networks that integrate kinds of renew- the HOMER tool was more apparent. Moreover, by increasing NPC
ables, etc. They are hot or fast developing processes that provide and CO2 emissions, a time-varying water profile had more
renewables to daily production or living activities and that also negative effects on the system performance compared to a static
attract lots of research interests. water profile.
Cambero et al. [94] pointed out that forest biomass (as a
5.1.1. Application in stand-alone energy systems renewable resource) had the potential to substitute fossil fuels in
As stochastic search algorithms are easier to deal with non- many applications. In the paper presented was a review of studies
convex and multimodal problems than gradient based classical that assessed or optimized economic, social and environmental
methods, heuristic algorithms like PSO and GA are used most in aspects of forest biomass supply chains for the production of
multi-objective optimization problems. However, not a single bioenergy and bioproducts. There was an inevitable trend that
optimization method can be said the best or worst based on one integrated these three aspects (measured respectively by pro-
application. For some applications one method may be better than duction and capital costs, the number of created jobs, and emis-
others. sions) into a multi-objective optimization problem. The optimal
Behera et al. [7] discussed the application of intelligent opti- design, planning and management of forest biomass supply
mization methods at various modules of a wind farm, in order to chains through multi-objective optimization approaches, based
maximize the whole power extraction from the wind. Optimization on techno-economic assessments and life cycle assessments
process in each module was undertaken separately, using separate (LCA), started to flourish and more practices were needed to
or hybrid intelligent optimization algorithms or combined with testify the performances.
classical methods.
For example, in the wind farm layout design process, the paper 5.1.2. Application in hybrid or integrated energy systems
combined intelligent optimization method and non-linear pro- Since energy crisis and greenhouse effect become increasingly
gramming method for this non-convex problem. The global optimal urgent issues all over the world, solar and wind energy come to play
solution was derived from the mathematical programming of the important roles in the energy structure with advantages of abun-
convex sub-region, and the initial solutions of the mathematical dant resources and little pollution. These clean and renewable
programming method were derived through intelligent algorithms. energy resources are also used to produce or replace or combine
In solving the classical Economic Load Dispatch (ELD) problem, the with conventional energy resources.
modified ICA was tested effective with short run time and fast In 2012, Fadaee and Radzi [97] made a review about using EAs
convergence performance. In the past, while solving the ELD for the multi-objective optimization of placement, sizing, design,
problem in the generation scheduling process, classical methods planning and control problems in hybrid renewable energy systems
like Lambda Iteration Method (LIM), gradient search method, linear (HERS). HERS was a green and reliable power system, especially for
programming, quadratic programming, dynamic programming and remote areas. Using MOEAs to reach the trade-offs between
Newton-based method, etc. were used [173]. However, ELD for different benefits and choosing the prior one based on preferences
large-scale systems often demonstrated as a highly nonlinear of decision-makers was realistic and difficult. Most engineering
constrained non-convex optimization problem, heuristic intelligent problems considered multiple objectives rather than single one,
algorithms such as EP, GA, GSA, DE, Adaptive PSO method, such as minimizing cost, maximizing performance, and maxi-
Sequential Quadratic Programming (SQP)-combined EP method, mizing reliability, etc. They found out that PSO, GA and their vari-
GA-SA combined method, etc. were applied [174e177]. ants were most popularly utilized algorithms in HERS which could
For further researches, Rodrigues et al. [93] made a deep dis- find global optima.
cussion on using EAs to solve a Multi-Objective Wind Farm Layout Efficient utilization of RESs is paid intense attention in order to
Optimization Problem (MOWFLOP), so as to realize trade-offs be- satisfy energy demands in various aspects of production processes
tween optimization objectives of energy production, capital in- and daily lives [179]. Zheng et al. [159] focused on the research of a
vestment and operational costs. Specifically in solving MOWFLOP, LSIES, which integrated distributed district heating and cooling
the work studied different characteristics of utilized multi- units (DHCs) and RESs (wind) through the power grid. They pro-
objective EAs (MOEAs, including NSGA-II) and introduced several posed a multi-objective group search optimizer method with
constraint-handling techniques. The relation between problem strategies of adaptive covariance [180] and Le vy flights [181]
dimensionality/complexity and layout design freedom was also (MGSO-ACL), for the power dispatch optimization of the LSIES.
studied, as well as the influence of problem size on algorithms The conflicting optimized objectives included the economic and
performance. As the state-of-the-art methods for solving multi- reliability interests of the power grid and DHCs. Experimental re-
objective optimization problems, several MOEAs were discussed sults demonstrated that the MGSO-ACL obtained superior Pareto-
and compared performances in solving MOWFLOP. The paper optimal solutions in terms of computation time, convergence, so-
presented a Multi-Objective Gene-pool Optimal Mixing Evolu- lution distribution and diversity with enhanced exploration and
tionary Algorithm (MOGOMEA), which outperformed classical exploitation abilities. It was also pointed out in the paper that the
NSGA-II for tested cases and reached trade-off solutions. proposed method was also applicable for various energy networks
Since the electrical grid is very costly for remote locations and including sub-systems, such as micro-grids and distributed energy
diesel power generation produces pollutants into the environ- units interconnected via power grids.
ment, RESs become more and more popular [178]. Clarke et al. For further researches, Wei et al. [161] proposed a multi-
[98] applied the PSO algorithm in the stand-alone renewable objective interval optimization model and ER method to solve the
energy system to achieve objectives of reducing the total net unit sizing optimization problem of small-scale integrated energy
present cost (NPC) and life-time CO2 emissions. The optimization systems. In the paper, the risks related to uncertain wind and solar
694 Y. Cui et al. / Energy 125 (2017) 681e704

energies were considered, together with the average life cycle cost generators played an important role in improving the accuracy and
(LCC). In order to solve this multi-objective optimization problem, a efficiency of distributed generation (DG) systems. In the paper, ob-
multi-objective group search optimizer method with adaptive jectives that were often selected to be optimized were described and
covariance and chaotic search (MGSOACC), which enhanced the explained, including technical objectives, financial objectives, and
adaptability and diversity of the optimizer, was developed. The combined multiple objectives. Optimization methods of different
proposed methods were tested on the direct district heating system categories were also presented solving different objectives. Whereas
(DH) and DHCs with superior search ability, achieving goals of low analytic methods or classical numerical methods were used with
cost, acceptable risk, and low pollutants emission. It was worth WSM to form a single objective optimization problem, intelligent
noticing that the variation of daily solar irradiation should be optimization algorithms that realized heuristic searches were more
handled carefully, whereas the peak-low-even method was used applicable for satisfying multiple objectives requirements, resulting
here to divide the daily heating and cooling loads into three in Pareto solutions. Hybrid intelligent algorithms were also intro-
representative periods. The proposed method can be applied to duced for solving the distributed generation optimization problem,
guide the investment behavior of the investors for trade-offs be- often with better results than single algorithms, especially for sys-
tween different interests. tems with integrated renewable resources.
Access to reliable electricity was necessary for improving the Mena et al. [182] introduced a multi-objective optimization
living standards. In consideration with assessing the convenience based on differential evolution (MOO-DE) framework for the
and appropriateness of a hybrid micro-grid system (HMGS) (of the integration of RESs into electric power networks. The framework
combination of wind, PV (photovoltaic), diesel generator, and bat- searched for the optimal size and location of different DG tech-
tery storage) [99], the trade-off between cost and reliability of the nologies, taking into account both uncertainties/risks related
system was the key to assess the system effectiveness and the to primary renewable resources availability, power demands,
quality of service. In the paper, a MultiObjective PSO (MOPSO) components failures, etc. and the economic concern. Whereas
method was used to find the best configuration of the system and uncertainties were measured by the index DCVaR(CG) (condi-
sizing of the components, helping to promote distribution of micro- tional value-at-risk deviation of global cost), the economic per-
grid projects and energy access especially in remote areas or formance was measured as the expected global cost (ECG). In the
developing countries. Since the energy resources were limited in paper, it also expected further extension of the optimization
different locations, the optimal ratings of wind, solar, battery, or framework, taking technical, social and geographical aspects (e.g.
diesel etc. differed with each other. The proposed method offered a the suitable geographical area for wind turbines and solar
promising way to overcome some technical barriers in the elec- photovoltaic) into considerations for trade-offs, too.
trification projects. It is obvious from previous researches that economic issues are
the foremost consideration in most production processes, not only
5.1.3. Application in distributed generation networks in energy (including RESs and traditional energy such as fuel, coal,
Niknama et al. [158] considered RESs (photovoltaics, fuel cell natural gas, etc.) production processes, but also in energy-
and wind energy) in the distribution network. They presented a consuming processes. At the same time, environmental issues
new multiobjective modified honey bee mating optimization and even social issues are included as optimization objectives in
(MHBMO) algorithm to realize optimization of total electrical en- latest years' researches. There are also continuous focuses on en-
ergy costs, electrical active power losses, the voltage deviations, ergy related issues such as energy efficiency (either production
and total emissions of the RESs and the grid. A set of non- efficiency or utilization efficiency), energy loss, energy system
dominated Pareto optimal solutions were stored in an external performance or reliability or stability, etc. Furthermore, perfor-
archive called repository, and a fuzzy clustering algorithm was used mance or technical objectives related to concrete processes or
to handle the varying size of the repository. Standard system tests systems may also be taken into account.
demonstrated superior exploration ability, feasibility and effec- The summary of application of multi-objective optimization
tiveness of the proposed algorithm. methods in renewable energy resources production processes is
As discussed in Ref. [96], different objectives, constraints, and given in Table 7.
optimization algorithms for the optimal allocation of distributed As can be seen from Table 7, EAs including PSO, GA, ABC, DE,

Table 7
Applications of multi-objective optimization methods in renewable energy resources production optimization.

Optimization Objectives (energy, economic, Optimization Methods Optimized System (renewable Authors
environment, technical, social aspects) energy resources)

energy production, capital investment, and EAs, a review wind farm layout optimization Rodrigues, Bauer, Bosman, 2016
operational costs problem
NPC and life-time CO2 emissions PSO stand-alone renewable energy Clarke, Al-Abdeli, Kothapalli, et al., 2015
system
production and capital costs, emissions, and EAs, a review forest biomass supply chains Cambero, Sowlati, 2014
the number of created jobs
energy reliability and economic interests MGSO-ACL; ER: decision-making LSIES Zheng, Chen, Wu, et al., 2015
method
renewable energy sources risks and average MGSOACC; ER small-scale integrated energy Wei, Wu, Jing, et al., 2016
LCC systems
renewable energy reliability and costs MOPSO hybrid micro-grid system Borhanazad, Mekhilef, Ganapathy et al., 2014
total electrical energy costs, total emissions, MHBMO integrated distribution network Niknama, Kavousifard, Tabatabaei, et al., 2011
electrical active power losses, and
voltage deviations
financial, technical objectives, a review intelligent optimization algorithms DG systems H.A., Huy, Ramachandaramurthy, et al., 2016
renewable energy sources uncertainties/ MOO-DE integrated electric power networks Mena, Hennebel, Li, et al., 2016
risks, the economic concern, power
demands, and components failures
 Y. Cui et al. / Energy 125 (2017) 681e704 695

COA, and their variants in improved forms are utilized in real- building performance was also analyzed. At last, the authors also
world renewable energy production processes for multi- looked forward to assessing uncertainty factors in the future.
objective optimization of economic, energetic, environmental or
technical issues. Actually, other intelligent algorithms can also be 5.2.2. Application in production processes
applied to these problems as long as they meet the demands of Because most production processes consume different kinds of
decision-makers in computation efficiency, solution precision or energy resources (both traditional and renewable energy re-
diversity, etc. Users can further improve intelligent algorithms sources), energy utilization optimization constitutes a crucial part
with different optimization techniques to make them more suit- of cost saving and environment protection.
able for solving concrete problems.
5.2.2.1. Application for renewable energy utilization optimization.
5.2. Application in energy saving optimization A systematic procedure based on the geographical information
system (GIS) for the multi-objective optimization of the energy
As mentioned above, energy and environment issues are also system was presented by Fazlollahi and Becker et al. [190]. The
crucial considerations in nowadays energy-consuming processes. systematic procedure included optimal process design and opera-
Except for cases considering only technical objectives that were tion of the energy distribution network and integration techniques
introduced in Chapter 3 and Chapter 4, we will present cases that for energy and resources transportation. Multi-objective optimi-
take energy and/or environment issues into account of the multi- zation of the energy system has been achieved through various
objective optimization and try to find trade-off optimal solution. optimization techniques, such as GA [191]. A multi-objective multi-
The energy issue, together with economic, environment or period optimization model consisting of process design and energy
technical issues, is important for sustainable development and integration techniques was described in the previous works
planet protection not only in production processes, but also in [192,193]. Then an improved multi-objective multi-period optimi-
living activities like building design or retrofitting. These processes zation methodology was developed, which was split up into four
aim at either improving the energy (utilization) efficiency or main stages: master optimization, thermo-economic simulation,
reducing the energy consumption. slave energy integration optimization and environ-economic
evaluation. In the master optimization stage, the evolutionary al-
5.2.1. Application in living activities gorithm was used to realize the simultaneous optimization of
In United States of the America (USA) in 2010, 41% of the total system efficiency, overall CO2 emissions and the total annual cost
primary energy and 74% of electricity is consumed by residential (TAC). In the slave energy integration optimization stage, the
and commercial buildings. Meanwhile, these buildings also pro- problem was concluded as a mixed integer linear model. The model
duce about 54% of SO2, 40% of CO2, and 17% of NOx emissions in determined the best resources schedule of selected subsystems to
energy consumption processes [183]. The current energy trends meet the requirements at a minimum cost, which was solved by
that raise great concerns are defined by the International Energy robust linear programming methods.
Agency as “three Es”: energy security, environment and economic Along with the rapid increase of world population and energy
prosperity [184]. Not only in USA and the European Union (EU) consumption, the effective utilization of renewable resources and
[185], but also around the whole world, people start to give priority low temperature waste heat has gained much attention. Feng et al.
to the optimization of energy consumption in building. Therefore, [108] used NSGA-II to realize thermo-economic multi-objective
in building design processes, the environmental effects as well as optimization, which involved both thermodynamic performance
energy efficiency are important considerations, too. Several studies and economic factors. The algorithm was applied in the optimiza-
have focused on developing design methods both energy-efficient tion processes of regenerative organic Rankine cycles (RORC) and
and environmental-friendly [186e188]. basic organic Rankine cycles (BORC), which were advanced tech-
Fesanghary et al. [189] aimed at the energy efficiency and nologies for converting low-grade heat resources into power [194].
environmental performance improvement of buildings. They used The optimization objectives were concluded as the maximization of
the EnergyPlus tool to model the whole energy system of the exergy efficiency [195] and the minimization of levelized energy
building, and applied HS to solve the optimal design problem. The cost (LEC) [196]. Based on the importance analysis of key operation
model aimed to realize energy consumption optimization, financial variables to optimization objectives, the bi-objective optimization
costs reduction and decrease of environmental impacts (repre- of RORC and BORC was conducted with favorable performances.
sented by LCC and CO2 equivalent emissions). Furthermore, the The results showed that both the exergy efficiency and the LEC of
proposed method was validated efficient on a typical single-family RORC were higher than those of BORC. It was because that the
house in the southern part of USA. Moreover, in the paper discussed higher exergy efficiency and thermodynamic efficiency led to lower
were a series of identified Pareto optimal solutions, which helped net power output and worse economic performance. On the basis
designers better understand the trade-off relation between eco- of engineering experiences, a decision making process was con-
nomic and environmental performances. ducted by a hypothetical point. In other words, the nearest point in
The retrofitting of buildings was important not only in the Pareto frontier to the ideal point was chosen as the desired and
improving occupants' comfort and health, but also in reducing the defined as the final Pareto-optimal solution. Furthermore, maxi-
energy consumption and green-house gasses emissions. Hence, mizing the net power output was added as the third optimization
Asadi et al. [103] combined GA with ANN to establish the multi- objective and the proposed algorithm was successful to find Pareto
objective optimization model for building retrofitting projects. optimal solutions.
Taking advantage of rapidity of ANN and optimization ability of GA, Naserian et al. [107] applied a controlled elitist NSGA-II to
conflicting objectives of energy consumption, retrofit cost and realize the multi-objective ecological optimization of a regenera-
thermal discomfort were optimized. They took a school building as tive Brayton cycle [197]. Based on finite-time thermodynamic
a case to take individual optimization of each objective, and then analysis [198] and exergy analysis in finite-size components [199],
simultaneous multi-objective optimization to survey the practica- the net output power and ecological states of the cycle were to be
bility and performance of the proposed method. During the ex- maximized, while the total exergy destruction was minimized by
periments, the competitive interaction between objectives and the introducing a dimensionless parameter, which embeds the time
influence of decision variables on each objective or even the variable. The utilized controlled elitist NSGA-II was tested with
696 Y. Cui et al. / Energy 125 (2017) 681e704

better convergence property and distribution of solutions than the trade-off solution between cost savings and thermodynamics effi-
original NSGA-II. The multi-objective optimization results provided ciency. LAPSO integrated with learning abilities through receiving
not one optimum point but a range of optimum ones, satisfying a favorable feedbacks from the environment gained extra flexibility
reliable and adoptable system even under variable daily power and exploitation abilities. In handling equality and inequality con-
demand. straints, an efficient and user-friendly feasibility based ranking
Solar energy, as a promising type of renewable resources, has strategy (FBRS) was applied. The numerical results showed that the
been utilized as energy input for Brayton type gas turbines for proposed algorithm was reliable with better solutions with
electricity generation [200,201]. Based on previous works robustness, compared with methods like GA or PSO. The proposed
[202,203], S anchez-Orgaz et al. [109] applied the NSGA-II to solve method was illustrated on a plate-fin heat exchanger but was
the multi-objective multi-parameter optimization problem of the applicable for all types of heat exchangers.
recuperative multi-step solar-driven Brayton thermo-solar plant. Based on previous researches [129,130,205], Geng et al. [100]
The optimized objectives involved the power output and the overall proposed an adaptive MOPSO (AMOPSO) algorithm for the opti-
thermodynamic efficiency of the thermo-solar plant, which were mization of the ethylene cracking furnace, referring to DEA (data
normalized by distance methods. Based on the thermo-dynamic envelopment analysis [206])-based energy efficiency analysis
analytical model of the plant, key design variables were recog- [207,208]. The optimization objectives involved maximizing prod-
nized. Moreover, several configurations including ideal, realistic, uct yields of ethylene and propylene, and minimizing consumption
single-stage or multi-step of the plant were analyzed and opti- of energies (i.e. the feedstock naphtha and steam). Besides, the
mized. The paper also applied different decision making ap- fuzzy consistent matrix method was used in the decision-making
proaches (i.e. Fuzzy, LINMAP, TOPSIS, Bellman-Zadeh, Ideal point process. The AMOPSO, compared with classical intelligent algo-
and Non-ideal point approaches) and made comparisons between rithms, produced superior convergent and distributed solutions
the optimal solutions. Seeing from the acceptable optimization with consistency in solving different problems. An ethylene
results physical insights were obtained by the multi-criteria deci- cracking furnace case was optimized using the proposed algorithm
sion method. to reach a viable balance between economic and energy consider-
Ju et al. [126] constructed a hybrid energy system based on a ations. Decision makers then chose the trade-off Pareto solution
combined cooling, heating and power (CCHP) system driven by catering to their preferences by fuzzy evaluation.
distributed energy resources (represented by natural gas, solar The summarization of application cases of multi-objective
energy and wind energy) (DERs CCHP). The optimized energy, optimization methods in energy saving related fields is given in
economic and environmental objectives of the energy system were Table 8.
respectively expressed as energy rate, total operation cost (TOC), It can be seen from Table 8 that EAs (like GA, PSO, HS and their
and carbon dioxide emission reductions. Experiments were con- variants, etc.) make good effect in multi-objective optimization of
ducted for mono-objective optimization of each optimization renewable and traditional energy production or utilization pro-
objective and multi-objective optimization of all three objectives cesses, taking into consideration of energy, economic, environment
based on the entropy weight method, comparatively on the hybrid or technical issues. Different processes make different models with
system and CCHP system driven solely by natural gas (NG CCHP) in various system features like non-linearity, multi-modal, and maybe
Guangzhou Higher Education Mega Center in China. The optimi- non-convexity. Making use of advantages of each intelligent opti-
zation results showed that the multi-objective optimization could mization algorithm with effective dealing strategies helps to solve
balance three optimization objectives and accommodate different different problems efficiently.
electric, cooling and heating loads. Moreover, the optimal solutions
and performances of DERs CCHP were better than those of NG 5.3. Application in emissions reduction optimization
CCHP.
In addition to economics, energy and reliability, environmental
5.2.2.2. Application for traditional energy utilization optimization. issues are also of great importance for the optimal design, con-
Shirazi et al. [162] made thermal (energetic and exergetic), eco- struction, production, use, or transportation of various activities
nomic, and environmental (emissions cost) analyses of an ITES around us people. Emissions (composed mainly of greenhouse
system for gas turbine cycle inlet air cooling. In the paper, a MOGA gases emissions) reduction is also linked closely with the protection
was employed to solve the simultaneous optimization problem of of our planet, both in living activities and production processes.
thermo-dynamic and economic objectives. The optimized objec-
tives were measured respectively by exergetic efficiency and the 5.3.1. Application in emissions reduction
total cost of the system, including the capital, maintenance and In order to minimize the operation cost and the environmental
operational costs (including the costs of electricity and fuel con- impact (gaseous emissions) of a micro-grid, Chaouachi et al. [95]
sumption) and the social cost of emissions. Moreover, the optimal formulated a multi-objective intelligent energy management
results achieved by utilizing the multi-objective method were framework (MIEM), using artificial intelligent algorithms together
compared with those by the single objective method considering with the linear-programming method. In dealing with constraints,
each of the mentioned objectives. The results demonstrated that the Artificial Neural Network Ensemble (ANNE) based on regular-
the multi-objective optimization provided a reasonable trade-off ized negative correlation learning (RNCL) was used to forecast
between objectives. There was another important factor for renewable energy resources generation (photovoltaic and wind
assessing the system to be covered: the payback period for the power generation) and load demand. A fuzzy logic expert system
installation investment. The additional costs would be compen- was used for battery scheduling with uncertainties, as a part of
sated over this period with the income received from the optimized online energy optimization. The proposed machine learning char-
system. acterized with enhanced learning and generation abilities was
Yousefi et al. [102] proposed a PSO algorithm based on learning successful in minimizing the operation cost and the emission level.
automata (LAPSO) to solve the design optimization problem of real- The framework was finally implemented on a micro-grid simula-
world compact heat exchangers (CHEs) [204] with varying heat tion model, and its effectiveness of solving competing objectives
duties. The LAPSO algorithm was implemented to deal with was validated.
changing design variables and operation conditions, so as to get the Zhang et al. [101] proposed a bare-bones multi-objective
 Y. Cui et al. / Energy 125 (2017) 681e704 697

Table 8
Applications of multi-objective optimization methods in energy saving.

Optimization Objectives (energy, economic, Optimization Methods Optimized System Authors

environment, technical, social aspects)

energy efficiency, LCC, and CO2 equivalent HS residential building Fesanghary, Asadi, Geem, 2012
emissions
energy consumption, retrofit cost and thermal GA with ANN building retrofitting projects Asadi, da Silva, Antunes, et al., 2014
discomfort
system efficiency, TAC, and overall CO2 EA energy distribution network cha, 2014
Fazlollahi, Becker, Mare
emissions
exergy efficiency, LEC, and the net power output NSGA-II RORC and BORC Feng, Zhang,Li, Shi, 2015
total exergy destruction, ecological states of the GA regenerative Brayton cycle Naserian, Farahat, Sarhaddi, 2015
cycle, and net output power
overall thermodynamic efficiency and power NSGA-II multi-step solar-driven Brayton plant S
anchez-Orgaz, Pedemonte, Ezzatti,
output et al., 2015
energy rate, total operation cost (TOC), and CO2 the entropy weight method, hybrid energy system based on CCHP Ju, Tan, Li, et al., 2016
emissions GAMS
exergetic efficiency and total cost(the capital, MOGA; TOPSIS: decision-making ITES Shirazi, Najafi, Aminyavari, et al., 2014
operational costs, and the social cost of method
emissions)
thermodynamics efficiency and cost savings LAPSO CHEs Yousefi, Darus, Yousefi, et al., 2015
naphtha and steam consumption, product AMOPSO ethylene cracking furnace Geng, Wang, Zhu, Han, 2016
yields

particle swarm optimization (BB-MOPSO) algorithm to solve the method based on NSGA-II was applied in a simulated hydro-treating
optimization problem of environmental/economic dispatch (EED). reactor to search for the optimal operational condition, which was
To handle the EED problem with conflicting objectives, many represented by minimal sulfur and aromatic compounds emissions
evolutionary algorithms were proposed [209e211], as well as the [106]. The simulation model was realized by a heterogeneous model
multi-objective mathematical programming method [212]. The of three main reactions: hydro-desulfurization, denitrogenation, and
variants of PSO such as the multi-objective chaotic PSO were dearomatization. The effects of key operational parameters such as
developed to solve the multi-objective EED optimization [213,214]. temperature, pressure, liquid hourly space velocity, and H2/oil ratio
The BB-MOPSO algorithm was distinctive in that it needed not tune on hydro-treating reactions were also evaluated, serving as guidance
up control parameters. Some other techniques such as constraint for the multi-objective optimization.
handling strategy, external archive storing elite particles and the
crowding distance method were used to make the BB-MOPSO al- 5.3.2. Application in ecological design
gorithm more effective in solving the multi-objective optimization Due to the growing awareness of the importance of protecting
of fuel cost and pollutants emissions. Moreover, the fuzzy mem- the environment, concepts like green design, ecological design,
bership function was used to present preferences of decision environmental benign design, cleaner production, etc. have been
makers to select the trade-off solution from the Pareto optimal proposed and used to realize the sustainable development of pro-
solutions. The simulation results showed that the proposed algo- duction processes [220]. Taghdisian et al. [125] proposed an
rithm efficiently produced well-distributed solutions approxi- ecological design method for sustainable development of methanol
mating the real Pareto set with good performance in a short production. In the paper, a green integrated methanol case (GIMC)
running time and it can be extended to other optimization prob- was presented and compared with the classical reference methanol
lems with good robustness. case (RMC). The environmental performances and operational
Water distribution networks are complex systems which variables of both cases were evaluated, using the life cycle assess-
require high investment in construction and maintenance. More- ment (LCA) method. CO2 was recognized as the main emitted
over, the optimal design problem of water distribution networks is pollutant in the methanol production (while other pollutants
a multi-objective difficult problem. Several works explored to solve involved CO, NOx and VOCs, etc.). Multiple objectives of maximizing
the problem in which the objectives are cost minimization and methanol production and minimizing CO2 emission in the GIMC
reliability maximization [215,216]. Herstein et al. [217] tried to were then optimized by GA to get optimal solutions with accept-
assess the environmental impact of water supply systems. Based on able level of diversity, offering economically and environmentally
the previous work which presented a methodology to evaluate the desirable design options. (LCA is a systematic methodology in
carbon emissions in the whole life cycle [218], Marques et al. [219] which the environmental impacts are considered in the product
combined a multi-objective SA algorithm with a hydraulic simu- total life cycle [221,222].)
lator to realize the optimal design and operation of water distri- For optimally designing the hybrid green power generation
bution networks. The combined method was used to optimize the systems (HGPS), Gharavi et al. [170] utilized the ICA to get the
total cost calculated as the sum of investment, operation and en- trade-off solution between economic and environmental objec-
ergy costs, environmental impacts as life cycle carbon emissions, tives, in consideration with reliability. The system was consisted of
energy consumption related emissions, and pressure violations. photovoltaic arrays, wind turbine units, fuel cell and electrolyzer.
Moreover, the decision tree was used here to describe the Pareto The ICA was verified as an algorithm with high convergence accu-
optimal set, assisting decision makers to select the optimal solution racy, and was appropriate for optimization of high-dimensional
according to certain preferences. The optimization results showed nonlinear HGPS with large number of variables and discontinuity
that the proposed method was able to deal with conflicting ob- features. The optimization results of non-autonomous HGPS were
jectives, environmental impacts and future uncertainty. also compared with those of autonomous HGPS.
The strict environmental regulations on sulfur and aromatic Fazlollahi and Mandel et al. [223] researched the optimal design
compounds emissions made the hydro-treating process of diesel fuel and operation of energy systems so as to reach the balance between
become more and more important. A multi-objective optimization energy supply and demand. They firstly presented a multi-period
698 Y. Cui et al. / Energy 125 (2017) 681e704

energy system optimization (ESO) model with a single objective farm in the Netherlands and produced a collection of alternative
function, making use of mixed integer linear programming (MILP) design options. It was also generic with the potential to support the
methods. Then, a multi-objective optimization method based on learning and decision-making processes of farmers, farm advisers
the evolutionary algorithm (EMOO) was presented. Integrating or scientists.
EMOO into the ESO model, a multi-objective investment and Schwartz et al. [104] pointed out that early design was shown to
operating optimization methodology was formed for the whole significantly impact the energy performance of buildings. The
complex energy system. The methodology included master and refurbishment of existing buildings was considered a cost-effective
slave optimization stages. In the master optimization stage, the way to the reduction of the life cycle environmental impact. Based
EMOO method was performed to optimize annual investment, on the LCA of the buildings, the LCCF (life cycle carbon footprint)
operation costs and CO2 emissions with a population of viable so- and the LCC over an assumed life span of 60 years were minimized.
lutions. Based on the integer cut constraints (ICC) and ε-constraint These two objectives were calculated based on the analysis of Life
methods concerning respectively integer and continuous variables, Cycle Energy use (LCE), consisted mainly of the embodied energy
the MILP was used in the slave optimization stage to optimize the required for construction, maintenance and refurbishments, and
mono objective of the total cost in short resolution time. It was operational energy used for maintaining the environmental and
pointed out in the paper that the resolution time of EMOO was thermal conditions within the building. Thus the LCCF was
possible to be improved via parallel computing. computed as the embodied costs and CO2 emissions, and the LCC as
Tamilselvi and Baskar [105] applied a covariance matrix adap- operational costs and CO2 emissions. The multi-objective optimi-
tation evolution strategy (CMA-ES) [224] to realize the optimal zation was realized by the computational efficient MOGA, which
transformer design (TD) problem [225], which was multi-modal, was one of the most widely used methods in the optimization of
multi-objective, mixed-variable and non-linear. The proposed building performances. Further, the CO2 payback time was calcu-
CMA-ES method was used to minimize individually mono objec- lated and the impact of building orientation on objectives was
tives of the purchase cost, the total life-time cost, the total mass and examined.
the total loss, in combination with penalty parameter less The summarization of application cases of multi-objective
constraint handling strategy. The optimal design results on three optimization methods in emissions reduction is listed in Table 9.
cases of CMA-ES were better than those of conventional design Adding or transferring to the environment objective, EAs are still
procedure, branch and bound algorithm, self-adaptive differential able to solve the multi-objective optimization problems in living
evolution or real coded genetic algorithm (RGA), either in global activities and production processes (including energy systems,
search ability, solution accuracy or consistency. Moreover, based on complex systems, agricultural and industrial production). Up till
the previous practices, NSGA-II was applied to realize multi- now, GA and PSO algorithms or their variants in improvement (like
objective TD optimization of the four objectives simultaneously to NSGA-II) are still popular in real-world applications. Intelligent
get a diverse set of solutions. The best trade-off solution chosen algorithms like ES, SA, ICA, etc. are also researched and applied with
then by fuzzy set theory was better than that obtained by CMA-ES satisfied optimization results. Nowadays, more and more algo-
in mono objective optimization. rithms, progressing with more powerful and applicable abilities,
Groot et al. [226] proposed a FarmDESIGN model to realize are put into use in realizing multi-objective optimization of real-
reconfiguration optimization of mixed farming systems, as well as world problems. Due to the limitation of the length of the paper,
meet farm and policy constraints, which was complicated with we could not cover all researches about different optimization is-
interrelated large array of components. The FarmDESIGN model sues in every field. However, various optimized objectives and
evaluated the productive, economic and environmental perfor- systems have more or less the same problem features that can be
mance of the farm by coupling a bio-economical farm model, and solved by choosing or improving effective intelligent algorithms.
generated a set of feasible non-dominated Pareto solutions.
Moreover, the Pareto-ranking and crowding distance methods
were used to select from feasible solutions. By using the Pareto- 6. Current difficulties and future directions for researches
based DE algorithm, objectives of maximizing operating profit
and organic matter balance, minimizing labor requirement and soil Although there are already many researches and continuous de-
nitrogen losses were optimized along the trade-off front. The pro- velopments about methods for solving multi-objective problems,
posed approach was finally testified on a realistic mixed organic there still exist difficulties to be solved or performance to be improved.

Table 9
Applications of multi-objective optimization methods in emissions reduction.

Optimization Objectives (environment, Optimization Method Optimized System Authors

economic, technical, social aspects)

emission level and operation cost artificial intelligent algorithms, micro-grid Chaouachi, Kamel, Andoulsi, et al., 2013
machine learning
pollutants emissions and fuel cost BB-MOPSO EED problem Zhang, Gong, Ding, 2012
life cycle carbon emissions, energy SA water distribution networks Marques, Cunha, Savi
c, 2015
consumption related emissions, total
cost, and pressure violations
sulfur and aromatic compounds NSGA-II simulated hydro-treating reactor Ani,Ebrahim, Azarhoosh, 2015
emissions
CO2 emission and methanol production GA, WPM GIMC Taghdisian, Pishvaie, Farhadi, 2015
environmental emission and economics ICA HGPS Gharavi, Ardehali, Ghanbari-Tichi, 2015
CO2 emissions, annual investment, and EMOO multi-period energy system Fazlollahi, Mandel, Becker, 2012
operation costs
total life-time cost, total mass, total loss, CMA-ES and NSGA-II distribution TD Tamilselvi, Baskar, 2014
and purchase cost
LCCF and LCC MOGA refurbishment of existing buildings Schwartz, Raslan, Mumovic, 2016
 Y. Cui et al. / Energy 125 (2017) 681e704 699

(1) The first-of-all important aim of multi-objective optimiza- objective, the optimization method is said to be effective and
tion methods is try to generate a set of well converged and the solution reasonable. The result comparison could be
uniformly distributed non-dominated optimal solutions. The made only between different algorithms or trade-off
convergence rates are also important. For different intelli- methods for the same process with same objectives, con-
gent algorithms, it is helpful to utilize different techniques in straints, etc. However, each attempt to utilize an intelligent
algorithms according to their distinct features. If we are able algorithm and trade-off method often takes considerable
to improve these algorithms from their intrinsic theorems time and efforts. How to select the optimization algorithm
and concepts, so as to make them more suitable for solving and trade-off method for concrete multi-objective optimi-
multi-objective optimization problems, the optimization zation problems often relies mostly on expert knowledge or
results will be more desirable. production experiences in this field and previous researches
(2) Unlike single-objective optimization, multi-objective opti- existed.
mization problems do not produce only one single solution, (2) Operational costs or investment or product profits are the
thus maintaining a diverse solution set keeps an important most obvious important concern in most production pro-
consideration. Diversity of non-dominated solutions pro- cesses and are relatively easy to be evaluated using LCA for
vides more selections for decision makers and helps to pre- example. Besides technical objectives that are related to each
vent from solution losses. It is worthy to be noticed that we distinct real-world optimization system or process, which is
should try to escape from local optimum solutions and find characterized with different features, energy considerations
global optima. In order to maintain the diversity and accel- (such as energy efficiency improvement or energy con-
erates the algorithm speed, the external archive is used by sumption reduction or energy saving potential, etc.) and
many algorithms. The elitism ensures that the best solutions environment considerations (i.e. greenhouse gases emis-
will not be lost. sions reduction, which is composed mainly of carbon diox-
(3) As an essential issue that is always under improvement, the ide) are gaining more and more attention. For energy
computation complexity remains to be reduced, computa- considerations, we need some support methods or tools to
tion time shortened and computation efficiency improved. quantify them, such as energy or exergy analysis, energy
On the other hand, the formulation of an efficient and balance or transfer equations expression, etc.
evidential stop criterion of multi-objective optimization (3) There are generally two ways to satisfy multiple objectives in
methods is difficult to realize. Judging the advance of opti- one optimization process: using the weight coefficients in
mization is as complex as the multi-objective optimization the WSM or WPM method or other a priori methods to
problem itself. Therefore, the optimization process stops transfer the multi-objective optimization problem into a
when it reaches a given number of iterations or function single objective optimization problem, which can then be
evaluations. solved by numerical or heuristic methods; applying appro-
(4) Most of the multi-objective optimization methods and the priate intelligent optimization algorithms to realize simul-
software tools are for analysis, guidance and planning, but in taneous optimization between objectives and decision
real-time application where the action should be taken making methods to achieve the trade-off. With different fo-
dynamically within seconds, the considerable computational cuses on optimized objectives, the optimization results offer
time is a disadvantage. The key of the dynamic or on-line differ with each other. By incorporating the local search
multi-objective optimization is the construction of discrete, technique into the optimization algorithm, we could achieve
equivalent and static sub-problems during each time period. fast convergence without loss in diversity. The comparison
How to express preferences of decision makers and share between independent optimization of each single objective
their information to appropriately steer the decision process and multi-objective optimization could offer some useful
is helpful in real-world optimization. Meanwhile, parallel advices for optimization potential of each objective.
computation will be appealing for lowering the computa- (4) When it comes to more than three objectives for most
tional burden. nonlinear and large-scale (and even non-convex) real-world
(5) When it comes to many-objective optimization problems optimization problems, the optimization processes become
with more than three objectives or high-dimensional opti- more complicated. Some techniques of dimensionality
mization problems, the optimization becomes more reduction, problem decomposition, constraint handling, and
complicated and the computation complexity increases convexification, etc. are utilized to simply the complex
exponentially. Decompose these problems effectively into optimization problem without losses of solutions as much as
sub-problems or use dimensionality reduction technique possible.
without losses of solutions make the key point, which is also
the difficult part. In the future, we will also carry out experiments on utilizing
some state-of-the-art tested effective intelligent optimization al-
As for the application in energy saving and environment pro- gorithms (such as NSGA-II, SPEA2, MOPSO, NNIA, and PESA-II, etc.)
tection fields, there are some concrete points need to be focused on: with trade-off methods in solving multi-objective optimization test
functions. We will further take the real-world ethylene cracking
(1) Because each multi-objective optimization problem in furnace as the example to verify the experiment results, with ob-
application is unique with distinct features, the optimized jectives of maximizing products yields of ethylene, propylene etc.,
objectives, applied intelligent algorithms and trade-off minimizing energy consumption and CO2 emissions, etc.
methods for decision making, constraints and handling
techniques are varied. As usual, there is no standard or 7. Conclusion
benchmark to evaluate the performance of application
optimization results. As long as the multi-objective optimi- Most real-world problems are intrinsically multi-objective
zation method finds the optimal solution set and the final optimization problems, including design, scheduling, modeling,
trade-off solution, satisfying all constraints and making a and optimization problems, etc. The optimized objectives range
compromise between objectives without harm to any one from materials or feedstock consumption, production, operation
700 Y. Cui et al. / Energy 125 (2017) 681e704

and maintenance cost, capital and investment, product yields, IH hypervolume evaluation index
product quality indexes, and profits, to energy efficiency or exergy IS spacing metric index
efficiency or energy consumption, to pollutant emissions (mainly Iε ε evaluation index
greenhouse gases emissions) such as CO2, SO2 emissions etc. One of m number of objectives
the biggest difficulties in solving multi-objective optimization n number of decision variables
problems lies in the trade-off among different conflicting and N population size
synergetic objectives. When we find a solution that makes one N external population size
objective optimal, it may leads another objective or other objectives P real Pareto optimal solution set
to fall into bad results. Since the multi-objective problem was firstly p number of inequality constraints
proposed, more and more researchers have made efforts to solve q number of equality constraints
the problem with effective methods, producing acceptable optimal r number of real Pareto solutions
solutions. s number of searched solutions
In this paper, the description of multi-objective optimization x decision vector
problems and solutions definition is given in summary. Due to the y objective vector
complexity, orthodox and mostly nonlinearity of multi-objective xi the ith decision variable
optimization problem, intelligent optimization algorithms like yi the ith objective
evolution based and swarm based algorithms were proposed and
has been improved continuously for solving the problem with good Greek letter
performance. We also give a brief introduction of some most often ε set of upper bounds of estimated objective function
used intelligent optimization algorithms about their development values
processes and (dis)advantages. Some existing test problems con- εi upper bound of ith objective function value
sisted of mathematical functions are also demonstrated and the wi a set of weighted coefficients
relative performance indexes for verifying the effectiveness of Q parameter space of decision vector
multi-objective optimization methods are summarized. J objective space
In order to get a final optimal solution for specific multi- U feasible domain of all feasible solutions
objective problems, trade-off optimization methods including a
prior methods, interactive methods, Pareto-based methods and
References
new dominance methods were proposed and improved. In the
Chapter 4, details about these trade-off methods are described. [1] Xu H. Research on multiobjective particle swarm optimization algorithms.
Meanwhile, applications of these trade-off methods in solving Shanghai: Shanghai Jiao Tong University; 2013.
multi-objective optimization problems are also taken as examples. [2] Gong M, Jiao L, Yang D, Ma W. Research on evolutionary multi-objective
optimization algorithms. J Softw 2009;20(2):271e89.
At last, applications of multi-objective optimization methods in [3] Miettinen K. Nonlinear multiobjective optimization. Boston: Kluwer Aca-
solving real-world multi-objective optimization problems, which demic Publishers; 1999.
aimed at multi-objective optimization of economic, energetic, [4] Marler R, Arora J. Survey of multi-objective optimization methods for engi-
neering. Struct Multidisc Optim 2004;26:369e95.
environmental, technical, and social concerns, are demonstrated. [5] Koopmans TC. Analysis of production as an efficient combination of activ-
The real-world problems include both energy (especially RESs) ities. In: Koopmans TC, Alchian A, Dantizg GB, Georgescu-Roegen N,
production and energy consumption processes. These applications Samuelson PA, Tucker AW, editors. Activity analysis of production and
allocation proceedings of a conference, vol. 13. New York: John Wiley and
can serve as guidance for solving multi-objective optimization Sons; 1951. p. 33e97.
problems in real-world production processes, making use of [6] Xu B. Research and application of multi-objective optimization algorithms
appropriate optimization algorithms and trade-off methods for based on differential evolution. Shanghai: China University of Science and
Technology; 2013.
simultaneous optimization of costs, energy efficiency or energy
[7] Behera S, Sahoo S, Pati BB. A review on optimization algorithms and appli-
consumption, environmental emissions, technical aspects or social cation to wind energy integration to grid. Renew Sustain Energy Rev
effects. 2015;48:214e27.
[8] Yang X. Biology-derived algorithms in engineering optimization. In: Olarius,
Zomaya, editors. Handbook of bio-inspired algorithms and applications. CRC
Acknowledgement Computer & Information Science Series: Chapman & Hall; 2006. p. 585e96.
[9] Neri F, Carlos C. Memetic algorithms and memetic computing optimization: a
literature review. Swarm Evol Comput 2012;2:1e14.
This work is partly financial supported by National Natural [10] Fogel LJ, Owens AJ, Walsh MJ. Intelligent decision making through a simu-
Science Foundation of China (61533003, 61603025), Natural Sci- lated evolution. Syst Res Behav Sci 1966;11:253e72.
[11] Hansen N, Arnold DV, Auger A. Evolution strategies. Springer handbook of
ence Foundation of Beijing, China (4162045) and National Key computational intelligence. 2015. p. 871e98.
Technology Support Program (2015BAK36B04). [12] Ba€ck T, Schwefel HP. An overview of evolutionary algorithms for parameter
optimization. Evol Comput 1993;1(1):1e23.
[13] Holland JH. Adaptation in natural and artificial systems. Michigan: The
Nomenclature University of Michigan Press; 1975.
[14] Schaffer JD. Multiple objective optimization with vector evaluated genetic
algorithms. In: Grefenstette JJ, editor. Proc. of the Int'l Conf. on Genetic Al-
A approximate Pareto optimal solution set gorithms and Their Applications. Hillsdale: L. Erlbaum Associates, Inc.; 1985.
di normalization Euclidean Distance between each searched p. 93e100.
solution and the real Pareto optimal solution [15] Fonseca CM, Fleming PJ. Genetic algorithm for multiobjective optimization:
formulation, discussion and generation. In: Forrest S, editor. Proc. of the 5th
doi desired value of each objective Int'l Conf. on Genetic Algorithms. San Mateo: Morgan Kauffman Publishers;
fi objective function 1993. p. 416e23.
[16] Srinivas N, Deb K. Multiobjective optimization using nondominated sorting
fk normalized objective function
in genetic algorithms. Evol Comput 1994;2(3):221e48.
G number of generations [17] Horn J, Nafpliotis N, Goldberg DE. A niched Pareto genetic algorithm for
IC convergence metric index multiobjective optimization. In: Fogaty TC, editor. Proc. of the 1st IEEE
ICS coverage of two sets Congress on Evolutionary Computation. Piscataway: IEEE; 1994. p. 82e7.
[18] Zitzler E, Thiele L. Multi-Objective evolution algorithm: a comparative case
IER error ratio study and the strength Pareto approach. IEEE Trans Evol Comput 1999;3(4):
IGD general distance 257e71.
 Y. Cui et al. / Energy 125 (2017) 681e704 701

[19] Zitzler E, Laumanns M, Thiele L. SPEA2: improving the strength Pareto 2010;1(1):1e16.
evolutionary algorithm. In: Giannakoglou K, Tsahalis DT, Periaux J, pez S, Portilla-Figueras A. The
[50] Salcedo-Sanz S, Del Ser J, Landa-Torres I, Gil-Lo
Papailiou KD, Fogarty T, editors. Evolutionary methods for design, optimi- coral reefs optimization algorithm: a novel meta-heuristic for solving hard
zation and control with applications to industrial problems. Berlin: Springer- optimization problems. Sci World J 2014;2014:1e15.
Verlag; 2002. p. 95e100. [51] Huynh TH. A modified shuffled frog leaping algorithm for optimal tuning of
[20] Knowles JD, Corne DW. Approximating the non-dominated front using the multivariable PID controllers. In: Proceedings of IEEE international confer-
Pareto archived evolution strategy. Evol Comput 2000;8(2):149e72. ence on industrial technology. IEEE ICIT; 2008. p. 1e6.
[21] Corne DW, Knowles JD, Oates MJ. The Pareto-envelope based selection al- [52] Duan H, Qiao P. Pigeon-inspired optimization: a new swarm intelligence
gorithm for multi-objective optimization. In: Schoenauer M, Deb K, optimizer for air robot path planning. Int J Intell Comput Cybern 2014;7(1):
Rudolph G, Yao X, Lutton E, Merelo JJ, et al., editors. Parallel problem solving 24e37.
from nature, PPSN VI. LNCS. Berlin: Springer-Verlag; 2000. p. 869e78. [53] Langton CG, Shimohara K. Artificial life five. The MIT Press; 1997.
[22] Corne DW, Jerram NR, Knowles JD, Oates MJ. PESA-II: region-Based selection [54] Adami C. Introduction to artificial life. Telos Pr; 1998.
in evolutionary multi-objective optimization. In: Spector L, Goodman ED, [55] Bonabeau E, Dorigo M, Theraulaz G. Swarm intelligence: from natural to
Wu A, Langdon WB, Voigt HM, Gen M, editors. Proc. of the Genetic and artificial systems. New York: Oxford University Press; 1999.
Evolutionary Computation Conf., GECCO 2001. San Francisco: Morgan [56] Millonas MM. Swarms, phase transitions, and collective intelligence. 1993.
Kaufmann Publishers; 2001. p. 283e90. arXiv preprint adap-org/9306002.
[23] Erickson M, Mayer A, Horn J. The niched Pareto genetic algorithm 2 applied [57] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of IEEE
to the design of groundwater remediation system. In: Zitzler E, Deb K, international conference on neural networks; 1995. p. 1942e8. 4(2).
Thiele L, Coello Coello CA, Corne D, editors. Proc. of the 1st Int'l Conf. on [58] Eberhart R, Kennedy J. A new optimizer using particle swarm theory. In:
Evolutionary Multi-Criterion Optimization, EMO 2001. Berlin: Springer- Sixth international symposium on micro machine and human science; 1995.
Verlag; 2001. p. 682e95. p. 39e43.
[24] Coello Ca Coello, Pulido GT. A micro-genetic algorithm for multiobjective [59] Li X. A non-dominated sorting particle swarm optimizer for multiobjective
optimization. In: Spector L, Goodman ED, Wu A, Langdon WB, Voigt HM, optimization. In: Cantu-Paz E, Foster JA, Deb K, Lawrence D, Roy R, editors.
Gen M, editors. Proc. of the Genetic and Evolutionary Computation Conf., Proc. of the Genetic and Evolutionary Computation Conf., GECCO 2003.
GECCO 2001. San Francisco: Morgan Kaufmann Publishers; 2001. p. 274e82. Springer-Verlag Berlin Heidelberg; 2003. p. 37e48.
[25] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist mulbi-objective [60] Coello Coello CA, Pulido GT, Lechuga MS. Handling multiple objectives with
genetic algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6(2):182e97. particle swarm optimization. IEEE Trans Evol Comput 2004;8(3):256e79.
[26] Jensen MT. Reducing the run-time complexity of multiobjective EAs: the [61] Rao RV, Savsani VJ, Vakharia DP. Teaching-learning-based optimization: a
NSGA-II and other algorithms. IEEE Trans Evol Comput 2003;7(5):503e15. novel method for constrained mechanical design optimization problems.
[27] Storn R, Price K. Differential Evolution-A simple and efficient adaptive Comput Aided Des 2011;43:303e15.
scheme for global optimization over continuous spaces. Berkeley, CA: In- [62] Hosseinpour H, Niknam T, Taheri SI. A modified TLBO algorithm for place-
ternational Computer Science Institute; 1995. ment of AVRs considering DGs. In: Proceedings of the 26th international
[28] Liu J, Lampinen J. A fuzzy adaptive differential evolution algorithm. Soft power system conference; 2011. p. 1e8.
Comput 2005;9(6):448e62. [63] Niknam T, Fard AK, Baziar A. Multi-objective stochastic distribution feeder
[29] Qin AK, Huang VL, Suganthan PN. Differential evolution algorithm with reconfiguration problem considering hydrogen and thermal energy pro-
strategy adaptation for global numerical optimization. IEEE Trans Evol duction by fuel cell power plants. Energy 2012;42:563e73.
Comput 2009;13(2):398e417. [64] Martín García JA, Gil Mena AJ. Optimal distributed generation location and
[30] Huang VL, Zhao SZ, Mallipeddi R, Suganthan PN. Multi-objective optimiza- size using a modified teachingelearning based optimization algorithm. Int J
tion using self-adaptive differential evolution algorithm. IEEE Congr Evol Electr Power Energy Syst 2013;50:65e75.
Comput 2009:190e4. CEC 2009; 2009. [65] Dasgupta D. An overview of artificial immune systems and their applications.
[31] Jia D, Zheng G, Khan MK. An effective memetic differential evolution algo- Springer-Verlag Berlin Heidelberg; 1998.
rithm based on chaotic local search. Inf Sci 2011;181(5):3175e87. [66] Coello Coello CA, Cortes NC. Solving multiobjective optimization problems
[32] Liu H, Cai Z, Wang Y. Hybridizing particle swarm optimization with differ- using an artificial immune system. Genet Program Evolvable Mach
ential evolution for constrained numerical and engineering optimization. 2005;6(2):163e90.
Appl Soft Comput 2010;10(2):629e40. [67] Gong M, Jiao L, Du H, Bo L. Multiobjective immune algorithm with non-
[33] P
aun G. Computing with membranes. J Comput Syst Sci 2000;61:108e43. dominated neighbor-based selection. Evol Comput 2008;16(2):225e55.
[34] P
aun G, Perez-Jimenez MJ. Membrane computing: brief introduction, recent [68] Shah-Hosseini H. Optimization with the nature-inspired intelligent water
results and applications. BioSystems 2006;85:11e22. drops algorithm. Int J Intell Comput Cybern 2008;1(2):193e212.
[35] P
aun G. A quick introduction to membrane computing. J Log Algebraic Pro- [69] Tayarani NMH, Akbarzadeh-T MR. Magnetic optimization algorithms a new
gram 2010;79:291e4. synthesis. In: Proceedings of IEEE congress on evolutionary computation,
[36] Leporati A, Zandron C, Ferretti C, Mauri G. Solving numerical NP-complete CEC 2008; 2008. p. 2659e64.
problems with spiking neural P systems. In: Eleftherakis G, et al., editors. [70] Rashedi E, Nezamabadi-pour H, Saryazdi SGSA. A gravitational search algo-
WMC8 2007, LNCS 4860. Springer-Verlag Berlin Heidelberg; 2007. rithm. Inf Sci 2009;179:2232e48.
p. 336e52. [71] Rashedi E, Nezamabadi-Pour H, Saryazdi S. Filter modeling using gravita-
[37] Liu W, Du Y. A novel focused crawler based on cell-like membrane tional search algorithm. Eng Appl Artif Intell 2011;24(1):117e22.
computing optimization algorithm. Neurocomputing 2014;123:266e80. [72] Kirkpatrick S, Gelatt CD, Vecchi MP. Optimization by simulated annealing. Sci
[38] Han M, Liu C, Xing J. An evolutionary membrane algorithm for global nu- J 1983;220(4598):671e80.
merical optimization problems. Inf Sci 2014;276:219e41. [73] Bandyopadhyay S, Saha S, Maulik U, Deb K. A simulated annealing based
[39] Singh G, Deep K, Nagar AK. Cell-like P-systems based on rules of particle multiobjective optimization algorithm: AMOSA. IEEE Trans Evol Comput
swarm optimization. Appl Math Comput 2014;246:546e60. 2008;12(3):269e83.
[40] Zhang G, Gheorghe M, Pan L, Pe rez-Jimenez MJ. Evolutionary membrane [74] Hamzacebi C, Kutay F. Continuous functions minimization by dynamic
computing: a comprehensive survey and new results. Inf Sci 2014;279: random search technique. Appl Math Modell 2007;31:2189e98.
528e51. [75] Velasquez Henao JD. An introduction to chaos-based algorithms for nu-
[41] Peng H, Luo X, Gao Z, Wang J, Pei Z. A novel clustering algorithm inspired by merical optimization. Rev Av Sist Inf 2011;8(1):51e60.
membrane computing. Hindawi Publ Corp Sci World J 2015;2015:8. Article [76] Huang H, Chung CY. Coordinated damping control design for DFIG-based
ID 929471. wind generation considering power output variation. IEEE Trans Power
[42] Song B, Song T, Pan L. Time-free solution to SAT problem by P systems with Syst 2012;27(4):1916e25.
active membranes and standard cell division rules. Nat Comput 2015;14: [77] Li B, Jiang W. Chaos optimization method and its application. Control Theory
673e81. Appl 1997;14(4):613e5.
[43] Geem ZM. Music-inspired harmony search algorithm: theory and applica- [78] Qin H, Li X. A chaotic search method for global optimization on tent map.
tions. Springer-Verlag Berlin Heidelberg; 2009. Electr Mach Control 2004;8(1):67e70.
[44] Wang L, Li L. An effective differential harmony search algorithm for the [79] Shan L, Qiang H, Li J, Wang Z. Chaotic optimization algorithm based on Tent
solving non-convex economic load dispatch problems. Int J Electr Power map. Control Decis 2005;20(2):179e82.
Energy Syst 2013;44(1):832e43. [80] Xu N, Zhou S, Zhang H, Yu JA. Hybrid chaos optimization method and its
[45] Karaboga D, Basturk B. On the performance of artificial bee colony (ABC) application. Syst Eng Electron 2003;25(2):226e7.
algorithm. Appl Soft Comput 2008;8(1):687e97. [81] Yang DX, Li G, Cheng GD. On the efficiency of chaos optimization algorithms
[46] Dorigo M, Gambardella L. Ant colony system: a cooperative learning for global optimization. Chaos, Solit Fractals 2007;34:1366e75.
approach to the travel ing sales man problem. IEEE Trans Evol Comput [82] Yan XF, Chen DZ, Hu SX. Chaos-genetic algorithms for optimizing the
1997;1(1):53e66. operating conditions based on RBF-PLS model. Comput Chem Eng
[47] Yang XS, Deb S. Engineering optimisation by cuckoo search. Int J Math Model 2003;27(10):1393e404.
Numer Optim 2010;1(4):330e43. [83] Gargari EA, Lucas C. Imperialist competitive algorithm: an algorithm for
[48] Yang XS. Firefly algorithms for multimodal optimization. In: Proceedings of optimization inspired by imperialistic competition. In: Proceedings of IEEE
the stochastic algorithms: foundation and applications, SAGA, vol. 5792. congress on evolutionary computation; 2007. p. 4661e7.
Springer-Verlag Berlin Heidelberg; 2009. p. 169e78. [84] Glover F. Heuristics for integer programming using surrogate constraints.
[49] Passino KM. Bacterial foraging optimization. Int J Swarm Intell Res Decis Sci 1977;8:156e66.
702 Y. Cui et al. / Energy 125 (2017) 681e704

[85] Glover F, Taillard E, de Werra D. A user's guide to tabu search. Ann Oper Res [112] Deb K, Pratap A, Meyarivan T. Constrained test problems for multi-objective
1993;41(1):3e28. evolutionary optimization: evolutionary multi-criterion optimization. In:
[86] Moscato P. On evolution, search, optimization, genetic algorithms and Lecture notes in computer science; 2001. p. 284e98.
martial arts: towards memetic algorithms. Pasadena, California, USA: Tech. [113] Huband S, Hington P, Barone L, While L. A review of multiobjective test
Rep. Caltech Concurrent Computation Program, Report 826, California problems and a scalable test problem toolkit. IEEE Trans Evol Comput
Institute of Technology; 1989. 2006;10(5):477e506.
[87] Reynolds G. An introduction to cultural algorithms. In: Proc of the annual [114] Li H, Zhang QF. Multiobjective optimization problems with complicated
conf on evolutionary programming. River Edge, NJ: World Scientific Pub- Pareto sets, MOEA/D and NSGA-II. IEEE Trans Evol Comput 2009;13(2):
lishing; 1994. p. 131e9. 284e302.
[88] Xie XF, Zhang WJ. Social cognitive optimization for nonlinear programming [115] Schaffer JD. Multiple objective optimization with vector evaluated genetic
problems. In: Proc of Int Conf Machine Learning and Cybernetics. Piscataway, algorithms. In: Proceedings of the 1st international conference on genetic
NJ: IEEE; 2002. p. 779e83. algorithms. Hillsdale, NJ, USA: L. Erlbaum Associates Inc.; 1985. p. 93e100.
[89] Corno F, Reorda MS, Squillero G. The selfish gene algorithm: a new evolu- [116] Fonseca C, Fleming P. Multiobjective optimization and multiple constraint
tionary optimization strategy. In: Proc of annual ACM symp on applied handling with evolutionary algorithms. ii. Application example. IEEE Trans
computing. Atlanta, USA: ACM; 1998. p. 349e55. Syst Man Cybern A Syst Hum 1998;28(1):38e47.
[90] Lin TY. Granular computing on binary relationsⅡ: rough set representations [117] Kursawe F. A variant of evolution strategies for vector optimization. In:
and belief functions. In: Rough sets in knowledge discovery. Heidelberg: Parallel problem solving from nature. 1st workshop, PPSN I, Lecture notes in
Physics-Verlag; 1998. p. 121e40. computer science, vol. 496. Springer-Verlag; 1991. p. 193e7.
[91] Picard RW. Affective computing. Cambridge. MA: MIT Press; 1997. [118] Deb K, Jain S. Running performance metrics for evolutionary multi-objective
[92] Zhou A, Qu BY, Li H, Zhao SZ, Suganthan PN, Zhang QF. Multiobjective optimization. Technical Report, No.2002004. Kanpur: Indian Institute of
evolutionary algorithms: a survey of the state of the art. Swarm Evol Comput Technology Kanpur; 2002. p. 13e20.
2011;1:32e49. [119] Schott JR. Fault tolerant design using single and multicriteria genetic algo-
[93] Rodrigues S, Bauer P, Bosman Peter AN. Multi-objective optimization of wind rithm optimization. Cambridge: Massachusetts Institute of Technology;
farm layoutseComplexity, constraint handling and scalability. Renew Sus- 1995.
tain Energy Rev 2016;65:587e609. [120] Coello CAC, Veldhuizen DAV, Lamont GB. Evolutionary algorithms for solving
[94] Cambero C, Sowlati T. Assessment and optimization of forest biomass supply multi-objective problems. New York: Kluwer Academic Publishers; 2002.
chains from economic, social and environmental perspectiveseA review of [121] Zitzler E, Thiele L. Multiobjective evolutionary algorithms: a comparative
literature. Renew Sustain Energy Rev 2014;36:62e73. case study and the Strength Pareto approach. IEEE Trans Evol Comput
[95] Chaouachi A, Kamel RM, Andoulsi R, Nagasaka K. Multiobjective intelligent 1999;3(4):257e71.
energy management for a microgrid. IEEE Trans Ind Electron 2013;60(4): [122] Veldhuizen DAV, Lamont GB. Multiobjective evolutionary algorithm
1688e99. research: a history and analysis. TR-98e03. Air Force Institute of technology;
[96] MP HA, Huy PD, Ramachandaramurthy VK. A review of the optimal alloca- 1998.
tion of distributed generation: objectives, constraints, methods, and algo- [123] Zitzler E, Thiele L, Laumanns M, Fonseca CM. Fonseca VGd. Performance
rithms. Renew Sustain Energy Rev 2016. http://dx.doi.org/10.1016/j.rser. assessment of multiobjective optimizers an analysis and review. IEEE Trans
2016.10.071. Evol Comput 2003;11(7):617e29.
[97] Fadaee M, Radzi MAM. Multi-objective optimization of a stand-alone hybrid [124] Xia L, Cui Y, Gu X, Geng Z. Kinetics modelling of ethylene cracking furnace
renewable energy system by using evolutionary algorithms: a review. Renew based on SQP-CPSO algorithm. Trans Inst Meas Control 2013;35(4):531e9.
Sustain Energy Rev 2012;16:3364e9. [125] Taghdisian H, Pishvaie MR, Farhadi F. Multi-objective optimization approach
[98] Clarke DP, Al-Abdeli YM, Kothapalli G. Multi-objective optimisation of for green design of methanol plant based on CO2-efficeincy indicator. J Clean
renewable hybrid energy systems with desalination. Energy 2015;88: Prod 2015;103:640e50.
457e68. [126] Ju L, Tan Z, Li H, Tan Q, Yu X, Song X. Multi-objective operation optimization
[99] Borhanazad H, Mekhilef S, Ganapathy VG, Modiri-Delshada M, Mirtaheri A. and evaluation model for CCHP and renewable energy based hybrid energy
Optimization of micro-grid system using MOPSO. Renew Energy 2014;71: system driven by distributed energy resources in China. Energy 2016;111:
295e306. 322e40.
[100] Geng Z, Wang Z, Zhu Q, Han Y. Multi-objective operation optimization of [127] Hananeh F, Alireza Z, Shahram J. The role of demand response in single and
ethylene cracking furnace based on AMOPSO algorithm. Chem Eng Sci multi-objective wind-thermal generation scheduling: a stochastic pro-
2016;153:21e33. gramming. Energy 2014;64:853e67.
[101] Zhang Y, Gong D, Ding Z. A bare-bones multi-objective particle swarm [128] Haimes Y, Lasdon L, Wismer D. On a bicriterion formulation of the problems
optimization algorithm for environmental/economic dispatch. Inf Sci of integrated system identification and system optimization. IEEE Trans Syst
2012;192:213e27. Man Cybernet 1971;3:296e7.
[102] Yousefi M, Darus AN, Yousefi M, Hooshyar D. Multi-stage thermal- [129] Geng Z, Cui Y, Xia L, Zhu Q, Gu X. Compromising adjustment solution of
economical optimization of compact heat exchangers: a new evolutionary- primary reaction coefficients in ethylene cracking furnace modeling. Chem
based design approach for real-world problems. Appl Therm Eng 2015;83: Eng Sci 2012;80:16e29.
71e80. [130] Cui Y. Ethylene cracking modeling and application research based on kinetics
[103] Asadi E, da Silva MG, Antunes CH, Dias L, Glicksman L. Multi-objective optimization. Beijing: Beijing University of Chemical Technology; 2012.
optimization for building retrofit: a model using genetic algorithm and [131] Wang F, Rao Y, Tang Q, He X, Zhang L. Fast construction method of Pareto
artificial neural network and an application. Energy Build 2014;81:444e56. non-dominated solution for multi-objective decision-making problem. Syst
[104] Schwartz Y, Raslan R, Mumovic D. Implementing multi objective genetic Eng Theory Pract 2015;35(1):1e10.
algorithm for life cycle carbon footprint and life cycle cost minimisation: a [132] Goldberg DE. Genetic algorithms in search, optimization and machine
building refurbishment case study. Energy 2016;97:58e68. learning. Boston, MA, USA: Addison- Wesley Longman Publishing Co., Inc.;
[105] Tamilselvi S, Baskar S. Modified parameter optimization of distribution 1989.
transformer design using covariance matrix adaptation evolution strategy. [133] Steuer RE, Choo E. An interactive weighted tchebycheff procedure for mul-
Electr Power Energy Syst 2014;61:208e18. tiple objective programming. Math Program 1983;26:326e44.
[106] Bakhshi Ani A, Ale Ebrahim H, Azarhoosh MJ. Simulation and Multi-Objective [134] Jaszkiewicz A, Słowinski R. The ‘Light Beam Search’ approach: an overview of
Optimization of a Trickle-Bed Reactor for diesel hydrotreating by a hetero- methodology and applications. Eur J Oper Res 1999;113:300e14.
geneous model using non-dominated sorting genetic algorithm II. Energy [135] Miettinen K, Ma €kela
€ M. Interactive bundle-based method for non-
Fuels 2015;29:3041e51. differentiable multiobjective optimization: NIMBUS. Optimization 1995;34:
[107] Naserian MM, Farahat S, Sarhaddi F. Finite time exergy analysis and multi- 231e46.
objective ecological optimization of a regenerative Brayton cycle consid- [136] Ojalehto V, Miettinen K, Laukkanen T. Implementation aspects of interactive
ering the impact of flow rate variations. Energy Convers Manag 2015;103: multiobjective optimization for modeling environments: the case of GAMS-
790e800. NIMBUS. Comput Optim Appl 2014;58(3):757e79.
[108] Feng Y, Zhang Y, Li B, Yang J, Shi Y. Comparison between regenerative [137] Hartikainen M, Miettinen K, Wiecek MM. Paint: Pareto front interpolation for
organic Rankine cycle (RORC) and basic organic Rankine cycle (BORC) based nonlinear multiobjective optimization. Comput Optim Appl 2012;52:
on thermoeconomic multi-objective optimization considering exergy effi- 845e67.
ciency and levelized energy cost (LEC). Energy Convers Manag 2015;96: [138] Wierzbicki AP. The use of reference objectives in multiobjective optimiza-
58e71. tion. In Multiple criteria decision making theory and application. In: Lecture
[109] Sanchez-Orgaz S, Pedemonte M, Ezzatti P, Curto-Risso PL, Medina A, notes in economics and mathematical systems, vol. 177. Berlin Heidelberg:
Herna ndez CA. Multi-objective optimization of a multi-step solar-driven Springer; 1980. p. 468e86.
Brayton plant. Energy Convers Manag 2015;99:346e58. [139] Das I, Dennis JE. Normal-boundary intersection: a new method for gener-
[110] Zitzler E, Deb K, Thiele L. Comparison of multi-objective evolutionary algo- ating the Pareto surface in nonlinear multicriteria optimization problems.
rithms: empirical results. Evol Comput 2000;8(2):173e95. SIAM J Optim 1998;8:631e57.
[111] Deb K, Thiele L, Laumanns M, Zitzler E. Scalable multi-objective optimization [140] Ganesan T, Elamvazuthi I, Shaari KZK, Vasant P. Swarm intelligence and
test problems. In: Fogel DB, editor. Proc. of the IEEE Congress on Evolu- gravitational search algorithm for multi-objective optimization of synthesis
tionary Computation, CEC 2002. Piscataway: IEEE Service Center; 2002. gas production. Appl Energy 2013;103:368e74.
p. 825e30. [141] Messac A, Ismail-Yahaya A, Mattson CA. The normalized normal constraint
 Y. Cui et al. / Energy 125 (2017) 681e704 703

method for generating the Pareto frontier. Struct Multidiscip Optim 2003;25: [168] Koduru P, Das S, Welch S. Multi-objective hybrid PSO using m-fuzzy domi-
86e98. nance. In: Genetic and evolutionary computation conference, GECCO 2007;
[142] Logist F, Van Impe J. Novel insights for multi-objective optimisation in en- 2007. p. 853e60.
gineering using normal boundary intersection and (enhanced) normalised [169] Kumar SS, Prasad YS. Modeling and optimization of multi objective non-
normal constraint. Struct Multidisc Optim 2012;45:417e31. linear programming problem in intuitionistic fuzzy environment. Appl
[143] Logist F, Van Erdeghem PMM, Van Impe JF. Efficient deterministic multiple Math Model 2015;39:4617e29.
objective optimal control of (bio)chemical processes. Chem Eng Sci 2009;64: [170] Gharavi H, Ardehali MM, Ghanbari-Tichi S. Imperial competitive algorithm
2527e38. optimization of fuzzy multi-objective design of a hybrid green power system
[144] YalçinKaya C, Maurer H. A numerical method for nonconvex multi-objective with considerations for economics, reliability, and environmental emissions.
optimal control problems. Comput Optim Appl 2014;57:685e702. Renew Energy 2015;78:427e37.
[145] Alvarez-V azquez L, García-Chan N, Martínez A, V azquez-Mendez M. Multi- [171] Zhang Q, Li H. MOEA/D: a multiobjective evolutionary algorithm based on
objective Pareto-optimal control: an application to wastewater manage- decomposition. IEEE Trans Evol Comput 2007;11(6):717e31.
ment. Comput Optim Appl 2010;46:135e57. [172] Mlakar M, Petelin D, Tusar T, Filipic B. GP-DEMO: differential evolution for
[146] Vallerio M, Vercammen D, Impe JV, Logist F. Interactive NBI and (E)NNC multiobjective optimization based on gaussian process models. Eur J Oper
methods for the progressive exploration of the criteria space in multi- Res 2015;243:347e61.
objective optimization and optimal control. Comput Chem Eng 2015;82: [173] Aravindhababu P, Nayar KR. Economic dispatch based on optimal lambda
186e201. using radial basis function network. Int J Electr Power Energy Syst
[147] Vallerio M, Hufkens J, Impe JV, Logist F. An interactive decision-support 2002;24(7):551e6.
system for multi-objective optimization of nonlinear dynamic processes [174] Swain RK, Sahu NC, Hota PK. Gravitational search algorithm for optimal
with uncertainty. Expert Syst Appl 2015;42:7710e31. economic dispatch. Proc Technol 2012;6:411e9.
[148] Miettinen K, Eskelinen P, Ruiz F, Luque M. NAUTILUS method: an interactive [175] Wang S-K, Chiou J-P, Liu CW. Non-smooth/non-convex economic dispatch
technique in multiobjective optimization based on the nadir point. Eur J by a novel hybrid differential evolution algorithm. IET Gener Transm Distrib
Oper Res 2010;206:426e34. 2007;1(5):793e803.
[149] Bortz M, Burger J, Asprion N, Blagov S, Bo € ttcher R, Nowak U, et al. Multi- [176] Panigrahi BK, Pandi VR, Das S. Adaptive particle swarm optimization
criteria optimization in chemical process design and decision support by approach for static and dynamic economic load dispatch. Energy Convers
navigation on Pareto sets. Comput Chem Eng 2014;60:354e63. Manag 2008;49(6):1407e15.
[150] Li Z, Nguyen TT, Chen S, Truong TK. A hybrid algorithm based on particle [177] Attaviriyanupap P, Kita H, Tanaka E, Hasegawa J. A hybrid EP and SQP for
swarm and chemical reaction optimization for multi-object problems. Appl dynamic economic dispatch with non-smooth fuel cost function. IEEE Trans
Soft Comput 2015;35:525e40. Power Syst 2002;17(2):411e6.
[151] Lam AYS, Li VOK. Chemical reaction-inspired metaheuristic for optimization. [178] Kornelakis A. Multiobjective particle swarm optimization for the optimal
IEEE Trans Evol Comput 2010;14(3):38e399. design of photovoltaic grid-connected systems. Sol Energy 2010;84:
[152] Shukla PK, Deb K. On finding multiple Pareto-optimal solutions using clas- 2022e33.
sical and evolutionary generating methods. Eur J Operational Res 2007;181: [179] Gebremedhin A. Introducing district heating in a Norwegian town-potential
1630e52. for reduced local and global emissions. Appl Energy 2012;95:300e4.
[153] Lu L, Cook ACM, Robinson TJ. A case study to demonstrate Pareto frontiers for [180] Hansen N, Müller SD, Koumoutsakos P. Reducing the time complexity of the
selecting a best response surface design with simultaneously optimizing derandomized evolution strategy with covariance matrix adaptation
multiple criteria. Appl Stoch Models Bus Ind 2012;3(28):85e96. (CMAES). Evol Comput 2003;11(1):1e18.
[154] Xiang Y, Zhou Y. A dynamic multi-colony artificial bee colony algorithm for [181] Reynolds AM, Smith AD, Reynolds DR, Carreck NL, Osborne JL. Honeybees
multi-objective optimization. Appl Soft Comput 2015;35:766e85. perform optimal scale-free searching flights when attempting to locate a
[155] Chen Z, Yuan Y, Yuan X, Huang Y, Li X, Li W. Application of multi-objective food source. J Exp Biol 2007;210(21):3763e70.
controller to optimal tuning of PID gains for a hydraulic turbine regulating [182] Mena R, Hennebel M, Li YF, Zio E. A multi-objective optimization framework
system using adaptive grid particle swam optimization. ISA Trans 2015;56: for risk-controlled integration of renewable generation into electric power
173e87. systems. Energy 2016;106:712e27.
[156] Huang L, Suh Il H, Abraham A. Dynamic multi-objective optimization based [183] DOE. Building energy data book. 2011 ed. US Department of Energy; 2012.
on membrane computing for control of time-varying unstable plants. Inf Sci [184] IEA-ISO. International Standards to develop and promote energy efficiency
2011;181:2370e91. and renewable energy sources. In: Special ISO focus. International Energy
[157] Yu K, Wang X, Wang Z. Self-adaptive multi-objective teaching-learning- Agency (IEA); 2007. p. 5e10.
based optimization and its application in ethylene cracking furnace opera- [185] EUROSTAT. Energy, transport and environment indicators. 2011 ed. EURO-
tion optimization. Chemom Intell Lab Syst 2015;146:198e210. STAT; 2011.
[158] Niknama T, Kavousifard A, Tabatabaei S, Aghaei J. Optimal operation man- [186] Diakaki C, Grigoroudis E, Kabelis N, Kolokotsa D, Kalaitzakis K, Stavrakakis G.
agement of fuel cell/wind/photovoltaic power sources connected to distri- A multi-objective decision model for the improvement of energy efficiency
bution networks. J Power Sources 2011;196:8881e96. in buildings. Energy 2010;35:5483e96.
[159] Zheng J, Chen J, Wu Q, Jing Z. Multi-objective optimization and decision [187] Gustavsson L, Joelsson A. Life cycle primary energy analysis of residential
making for power dispatch of a large-scale integrated energy system with buildings. Energ. Build 2010;42:210e20.
distributed DHCs embedded. Appl Energy 2015;154:369e79. [188] Asadi E, da Silva MG, Antunes CH, Dias L. A multi-objective optimization
[160] Chin KS, Fu C. Integrated evidential reasoning approach in the presence of model for building retrofit strategies using TRNSYS simulations. GenOpt
cardinal and ordinal preferences and its applications in software selection. MATLAB. Build Environ 2012;56:370e8.
Expert Syst Appl 2014;41(15):6718e27. [189] Fesanghary M, Asadi S, Geem ZW. Design of low-emission and energy-
[161] Wei F, Wu Q, Jing Z, Chen J, Zhou X. Optimal unit sizing for small-scale in- efficient residential buildings using a multi-objective optimization algo-
tegrated energy systems using multi-objective interval optimization and rithm. Build Environ 2012;49:245e50.
evidential reasoning approach. Energy 2016;111:933e46. chal F. Multi-objective, multi-period optimization
[190] Fazlollahi S, Becker G, Mare
[162] Shirazi A, Najafi B, Aminyavari M, Rinaldi F, Taylor RA. Thermal-economic- of district energy systems: III. distribution networks. Comput Chem Eng
environmental analysis and multi-objective optimization of an ice thermal 2014;66:82e97.
energy storage system for gas turbine cycle inlet air cooling. Energy 2014;69: [191] Chicco G, Mancarella P. Distributed multi-generation: a comprehensive view.
212e26. Renew Sustain Energy Rev 2009;13:535e51.
[163] Laumanns M, Thiele L, Deb K, Zitzler E. Combining convergence and diversity [192] Fazlollahi S, Bungener SL, Mandel P, Becker G, Mare chal F. Multi-objectives,
in evolutionary multi-objective optimization. Evol Comput 2002;10(3): multi-period optimization of district energy systems: I. Selection of typical
263e82. periods. Comput Chem Eng 2014;65:54e66.
[164] Hermandez-Diaz AG, Santana-Quintero LV, Coello Coello CA, Molina J. Par- [193] Fazlollahi S, Becker G, Mare chal F. Multi-objectives, multi-period optimiza-
eto-Adaptive ε-dominance. Evol Comput 2007;15(4):493e517. tion of district energy systems: II. Daily thermal storage. Comput Chem Eng
[165] Brockoff D, Zitzler E. Are all objective necessary on dimensionality reduction 2014;71:648e62.
in evolutionary multi-objective optimization. In: Runarsson TP, Beyer HG, [194] Hung TC, Shai TY, Wang SK. A review of organic Rankine cycles (ORCs) for
Burke E, Merelo-Guervos JJ, Whitley LD, Yao X, editors. Parallel problem the recovery of low-grade waste heat. Energy 1997;22:661e7.
solving from nature, PPSN IX. LNCS. Berlin: Springer-Verlag; 2006. [195] Roy JP, Mishra MK, Misra A. Performance analysis of an organic Rankine cycle
p. 533e42. with superheating under different heat source temperature conditions. Appl
[166] Deb K, Saxena DK. In: Deb Kalyanmoy, Saxena Dhish Kumar, editors. On Energy 2011;88:2995e3004.
finding Pareto-optimal solutions through dimensionality reduction for [196] Zhang SJ, Wang HX, Guo T. Performance comparison and parametric opti-
certain large-dimensional multi-objective optimization problems; 2005. mization of subcritical organic Rankine cycle (ORC) and transcritical power
p. 1e19. Techinical Report, 2005011, Kanpur: Indian Institute of Technology. cycle system for low-temperature geothermal power generation. Appl En-
[167] Saxena DK, Deb K. Non-linear dimensionality reduction procedure for ergy 2011;88:2740e54.
certain large-dimensional multi-objective optimization problems: employ- [197] Herrera CA, Sandoval JA, Rosillo ME. Power and entropy generation of an
ing correntropy and a novel maximum variance unfolding. In: Coello extended irreversible Brayton cycle: optimal parameters and performance.
Coello CA, Aguirre AH, Zitzler E, editors. Proc. of the 4th Int'l Conf. on J Phys D Appl Phys 2006;39:3414e24.
Evolutionary Multi-Criterion Optimization, EMO 2007. Berlin: Springer- [198] Sadatsakkak SA, Ahmadi MH, Bayat R, Pourkiaei SM, Feidt M. Optimization
Verlag; 2007. p. 772e87. density power and thermal efficiency of an endoreversible Braysson cycle by
704 Y. Cui et al. / Energy 125 (2017) 681e704

using non-dominated sorting genetic algorithm. Energy Convers Manage 1497e504.

2015;93:31e9. [213] Cai J, Ma X, Li Q, Li L, Peng H. A multi-objective chaotic particle swarm
[199] Ust Y, Sahin B, Kodal A, Akcay IH. Ecological coefficient of performance optimization for environmental/economic dispatch. Energy Convers Manag
analysis and optimization of an irreversible regenerative-Brayton heat en- 2009;50(5):1318e25.
gine. Appl Energy 2006;83(6):558e72. [214] Chen Y, Wang W. Environmentally constrained economic dispatch using
[200] Le Roux WG, Bello-Ochende T, Meyer JP. A review on the thermodynamic Pareto archive particle swarm optimization. Int J Syst Sci 2010;41(5):
optimization and model of the solar thermal Brayton cycle. Renew Sust 593e605.
Energy Rev 2013;28:677e90. [215] Di Pierro F, Khu ST, Savic D, Berardi L. Efficient multi-objective optimal
[201] S
anchez-Orgaz S, Medina A, Hern andez AC. Maximum overall efficiency for a design of water distribution networks on a budget of simulations using
solar-driven gas turbine power plants. Int J Energy Res 2013;37:1580e91. hybrid algorithms. Environ Model Softw 2009;24(2):202e13.
[202] Ahmadi MH, Ahmadi MA, Feidt M. Performance optimization of a solar- [216] Creaco E, Franchini M. Fast network multi-objective design algorithm com-
driven multi-step irreversible Brayton cycle based on a multi-objective ge- bined with an a posteriori procedure for reliability evaluation under various
netic algorithm. Oil Gas Sci Technol 2016;71(1):1e14. operational scenarios. Urban Water J 2012;9(6):385e99.
[203] S
anchez-Orgaz S, Medina A, Hern andez AC. Thermodynamic model and [217] Herstein L, Filion Y, Hall K. Evaluating the environmental impacts of water
optimization of a multi-step irreversible Brayton cycle. Energy Convers distribution systems by using EIO-LCA-based multiobjective optimization.
Manage 2010;51:2134e43. J Water Resour Plan Manag 2011;137(2):162e72.
[204] Li Q, Flamant G, Yuan X, Neveu P, Luo L. Compact heat exchangers: a review [218] Marques JCR, Cunha MC, Savic DA. Using real options for an eco-friendly
and future applications for a new generation of high temperature solar re- design of water distribution systems. J Hydroinformatics 2015;17(1):20e35.
ceivers. Renew Sustain Energy Rev 2011;15(9):4855e75. [219] Marques JCR, Cunha MC, Savi c DA. Multi-objective optimization of water
[205] Xia L, Chu J, Geng Z. Dynamic optimization based on adaptive particle swarm distribution systems based on a real options approach. Environ Model Softw
optimization algorithm. Comput Appl Chem 2012;136:133e40. 2015;63:1e13.
[206] Han Y, Geng Z, Liu Q. Energy efficiency evaluation based on data envelop- [220] Lim SR, Park JM. Environmental indicators for communication of life cycle
ment analysis integrated analytic hierarchy process in ethylene production. impact assessment results and their applications. J Environ Manage 2009;90:
Chin J Chem Eng 2014;22(12):1279e84. 3305e12.
[207] Han Y, Geng Z, Gu X, Zhu Q. Energy efficiency analysis based on DEA inte- [221] ISO. ISO 14040 international standard. Environmental management-life cy-
grated ISM: a case study for Chinese ethylene industries. Eng Appl Artif Intell cle assessment- principles and framework. Geneva, Switzerland: Interna-
2015;45:80e9. tional Organization for Standardization; 2006.
[208] Geng Z, Han Y, Gu X, Zhu Q. Energy efficiency estimation based on data [222] Finnveden G, Hauschild MZ, Ekvall T, Guinee J, Heijungs R, Hellweg S, et al.
fusion strategy: case study of ethylene product industry. Ind Eng.Chem Res Recent developments in life cycle assessment. J Environ Manag 2009;91:
2012;51:8526e34. 1e21.
[209] Abido MA. Multiobjective evolutionary algorithms for electric power [223] Fazlollahi S, Mandel P, Becker G, Mare chal F. Methods for multi-objective
dispatch problem. IEEE Trans Evol Comput 2006;10(3):315e29. investment and operating optimization of complex energy systems. Energy
[210] Basu M. Dynamic economic emission dispatch using nondominated sorting 2012;45(1):12e22.
genetic algorithm e II. Electr Power Energy Syst 2008;30(2):140e9. [224] Hansen N, Ostermeier A. Completely derandomized self-adaptation in evo-
[211] Bhattacharya A, Chattopadhyay PK. Application of biogeography-based lution strategies. Evol Comput 2001;9(2):159e95.
optimization for solving multi-objective economic emission load dispatch [225] Amoiralis EI, Tsili MA, Kladas AG. Transformer design and optimization: a
problems. Electr Power Compon Syst 2010;38(3):340e65. literature survey. IEEE Trans Power Deliv 2009;24(4):1999e2024.
[212] Vahidinasab V, Jadid S. Joint economic and emission dispatch in energy [226] Groot JCJ, Oomen GJM, Rossing WAH. Multi-objective optimization and
markets: a multiobjective mathematical programming. Energy 2010;35(3): design of farming systems. Agric Syst 2012;110:63e77.