An enhanced surrogate-assisted differential

evolution for constrained optimization problems

Rafael de Paula Garcia
Federal University of Vicosa: Universidade Federal de Vicosa
Beatriz Souza Leite Pires de Lima
Federal University of Rio de Janeiro: Universidade Federal do Rio de Janeiro
Afonso Celso de Castro Lemonge (  [email protected] )
Federal University of Juiz de Fora: Universidade Federal de Juiz de Fora https://orcid.org/0000-0001-
9938-294X
Breno Pinheiro Jacob
Federal University of Rio de Janeiro

Research Article

Keywords: Constrained optimization problems, Evolutionary algorithms, Surrogate models, Constraint-

handling techniques

DOI: https://doi.org/10.21203/rs.3.rs-522951/v1

License:   This work is licensed under a Creative Commons Attribution 4.0 International License.
Read Full License
 −1−

An enhanced surrogate-assisted differential evolution

for constrained optimization problems
Rafael de Paula Garciaa,1, [email protected]
Beatriz Souza Leite Pires de Limaa, [email protected]
Afonso Celso de Castro Lemongeb,*, [email protected]
Breno Pinheiro Jacoba, [email protected]
a
PEC/COPPE/UFRJ – Post-Graduate Institute of the Federal University of Rio de Janeiro, Civil Engineering Dept.
Avenida Pedro Calmon, S/N, Cidade Universitária, Ilha do Fundão
21941-596 Rio de Janeiro, RJ, Brazil.
http://www.lamcso.coppe.ufrj.br
b
UFJF – Federal University of Juiz de Fora, Applied and Computational Mechanics Dept.
Rua José Lourenço Kelmer, S/N
36036-330 Juiz de Fora, MG, Brazil
http://www.ufjf.br/mac/
*Corresponding Author.
1
Present address: UFV – Federal University of Viçosa, Department of Architecture and Urban Planning.
Avenida Peter Henry Rolfs, s/n, Campus Universitário
36570-900, Viçosa, MG, Brazil

Abstract
The application of Evolutionary Algorithms (EAs) to complex engineering
optimization problems may present difficulties as they require many evaluations of
the objective functions by computationally expensive simulation procedures. To deal
with this issue, surrogate models have been employed to replace those expensive
simulations. In this work, a surrogate-assisted evolutionary optimization procedure is
proposed. The procedure combines the Differential Evolution method with a 𝑘-
nearest neighbors (𝑘–NN) similarity-based surrogate model. In this approach, the
database that stores the solutions evaluated by the exact model, which are used to
approximate new solutions, is managed according to a merit scheme. Constraints are
handled by a rank-based technique that builds multiple separate queues based on the
values of the objective function and the violation of each constraint. Also, to avoid
premature convergence of the method, a strategy that triggers a random
reinitialization of the population is considered. The performance of the proposed
method is assessed by numerical experiments using 24 constrained benchmark
functions and 5 mechanical engineering problems. The results show that the method
achieves optimal solutions with a remarkably reduction in the number of function
evaluations compared to the literature.

Keywords: Constrained optimization problems; Evolutionary algorithms; Surrogate models;

Constraint-handling techniques
 −2−

1 Introduction
EAs have become appealing to solve complex real-world engineering optimization
problems, for which classical mathematical methods may no longer be effective. This includes,
for instance, the design of offshore oil & gas production systems such as floating platforms,
risers, mooring systems and submarine pipelines [1-3]: those problems are characterized by
many design variables, defined on high-dimensional spaces, with objectives and constraints that
are generally non-linear functions. However, the use of EAs for such problems can become
impractical due to excessive computational costs: many evaluations of candidate solutions may
be needed until a satisfactory solution is found, and each evaluation requires computationally
expensive numerical simulations.

Those issues can be handled by building cheaper approximation models (usually referred
as surrogate or metamodels) [4] to partially or totally replace the numerical simulation
procedures, emulating or mimicking the behavior of the actual model as closely as possible by
capturing the relationship between a given set of inputs and outputs. The goal is to significantly
reduce computational costs, while presenting accurate results. In this sense, several surrogate
models have been proposed. They may comprise polynomial approximation models [5,6]; radial
basis functions [7-10]; response surface models [11]; artificial neural networks [12]; Gaussian
process [13,14]; Kriging [15]; Support Vector Machines [16], and similarity-based models such
as the nearest-neighbor algorithms [17-19]. Particularly, the 𝒌-NN model presents many
advantages in terms of good performance and ease of implementation, as reported in [19-21].

Many applications of surrogate models with EAs have been presented in the literature. In
[22] a polynomial model was used to approximate the objective function. In [23] a kriging
metamodel, trained using few samples evaluated by the actual objective function, has been
successfully used to reduce the number of function evaluations. The goal of reducing
computational costs was also sought in [24] where the fitness function has been approximated by
an online multi-layer neural network, and in [25] that presented a surrogate-assisted PSO
algorithm. Surrogate models have also been employed to initialize the population of candidate
solutions in evolutionary operators such as mutation and crossover [26,27], and to calculate
approximations for the sum of violation terms [21]. The IEEE Congress on Evolutionary
Computation (CEC) competitions have presented many articles that solve computationally
expensive functions by employing metamodels with EAs. Regarding particularly the application
to offshore oil & gas production systems mentioned above, surrogate models based on neural
networks were used in [28,29] to avoid expensive nonlinear dynamic Finite Element analyses.

Among evolutionary algorithms, the Differential Evolution (DE) algorithm [30] has been
widely used due to its outstanding performance. The association of DE with approximation
models has been studied in some works, including neural networks [31] and neural networks of
radial basis functions [32]. In [33] a simple kriging metamodel was proposed to comprise a
surrogate model-based differential evolution (ESMDE). The proposed strategy was based on the
population members of the current generation, to assist the DE to generate competitive offspring
by using an appropriate parameter set along the different stages of the evolution.
 −3−

A surrogate model for DE consisting of multiple local surrogate models was presented in
[34], built from the feedback of the evolutionary process in diverse overlapped local regions of
the search space. In [35] a multi-fidelity surrogate-model-based DE optimization strategy was
developed, coupled to a data mining technique to handle the discrepancy between the low and
high fidelity simulation models. A surrogate-model to assist the DE algorithm in generating
competitive solutions along the evolutionary process was proposed in [36], using an adaptation
scheme to adapt parameters of a Kriging model. Recently, a two-layer adaptive surrogate-
assisted DE has been proposed in [37], where three different search strategies are adapted during
the evolutionary process according to the feedback information, to measure the status of the
algorithm approaching the optimal value. Local similarity-based surrogate models based on an r-
nearest neighbors have been employed in [17] and [18] to solve complex structural optimization
problems; the results indicated that the performance of DE has been significantly improved. In
this context, in [38] we had presented preliminary studies leading to the optimization procedure
called SADE-kNN (Surrogate-Assisted Differential Evolution with the 𝑘-NN surrogate model).
Now, this work presents and evaluates an improved surrogate-assisted evolutionary
method, which will be referred to as SADE-kNN-MCR, intended to comprise a more efficient
optimization tool to solve complex real-world engineering problems. This method introduces
enhanced techniques, beginning with the Multiple Constraint Ranking (MCR) constraint-
handling technique (CHT) [39]. The MCR comprises a rank-based approach that builds multiple
separate queues, corresponding to the values of the objective function and the violation of each
constraint. It has been devised to deal with complex engineering problems characterized by a
large number of constraints that may present different units and/or different orders of magnitude,
where the possible range of minimum and maximum violation values may not be known a-priori
(thus standard normalization procedures cannot be applied).
The proposed SADE-kNN-MCR method also incorporates a merit-based scheme to
manage the database used by the 𝒌-NN surrogate model to store the solutions evaluated by the
exact model, privileging solutions able to better approximate new candidate solutions. Moreover,
results of preliminary studies indicated that the previous SADE- 𝒌NN method presented by the
authors in [38] is susceptible to premature convergence and/or stagnation around local optima.
This is a known issue related to DE algorithms, as acknowledged in the literature [40,41]. To
circumvent this issue, a strategy that triggers a random reinitialization of the population
whenever it stagnates on local optima has been incorporated in the proposed method, comprising
the variant called SADE-kNN-MCR+.

The performance of the proposed method is compared with other surrogate-assisted

procedures recently presented in the literature, by applying them to several benchmark functions
(including the G-suite from the IEEE competition on real parameter constrained optimization
[42], and mechanical engineering problems), and assessing their relative performance by direct
comparisons and nonparametric statistical tests.
 −4−

This paper is organized as follows: Initially, Section 2 summarizes the Differential

Evolution algorithm, the 𝑘-NN similarity-based metamodel, and the CHTs employed in
surrogate-assisted evolutionary methods recently presented in the literature. Section 3 then
describes the proposed surrogate-assisted differential evolution methods. The full sets of
numerical experiments are described in Section 4. The results are presented in Sections 5 and 6.
The former compares the variants of the proposed method (SADE-kNN-MCR and SADE-kNN-
MCR+) with the SADE-kNN previously presented by the authors in [38]. Then, Section 6
presents comparisons with other methods recently presented in the literature that are listed in
Section 2.4. Final remarks and conclusions are presented in Section 7.

2 Related Works
2.1 Differential Evolution
The DE algorithm, proposed by Storn and Price in the nineties [30], has been shown to be
a very effective algorithm to solve complex real-world optimization problems such as system
design [43] and robot manipulator design [44]. One of the main characteristics of the DE is its
simplicity and ease of implementation. A detailed review of its basic concepts is provided by
[45,46], along with a survey describing its major variants, covering several aspects related to its
application to multiobjective, constrained, large scale and uncertain engineering optimization
problems.

Four DE variants or “strategies” were proposed in [47,48], characterized by the procedures

adopted for the selection of the individuals, and the type of recombination. In this work the first
one (DE/rand/1/bin) will be adopted, as follows:
𝑢𝑖,𝑗,𝐺+1 = 𝑥𝑟1,𝑗,𝐺 + 𝐹(𝑥𝑟2,𝑗,𝐺 − 𝑥𝑟3,𝑗,𝐺 ) (2.1)

where 𝐹 is the scale factor of the algorithm, and 𝑟1, 𝑟2 and 𝑟3 are randomly selected individuals.
A crossover operator CR is also defined, to combine the information of the solutions in the
current population and the solutions generated by Equation 2.1.

2.2 The k-NN similarity-based metamodel

As mentioned in the Introduction, due to this simplicity and good performance, the 𝑘-NN
approximation method has been selected in this work to comprise a surrogate model coupled to
the DE algorithm. In general, the 𝑘-NN method evaluates a solution by calculating the average
of the objective function of the 𝑘-nearest solutions weighted by its distances. This technique can
be classified in two categories:
• 𝑘-nearest neighbors (𝑘-NN) proposed in [49], where the 𝑘 nearest candidate solutions are
able to be selected; and
• 𝑟-nearest neighbors (𝑟-NN) proposed in [50], for a given neighborhood controlled by 𝑟
defines the candidate solutions able for selection.
 −5−

In this paper, the k-NN similarity-based metamodel is adopted. The solutions (𝑥, 𝑓(𝑥))
evaluated by the exact function are stored in a database as an ordered set 𝐷. Any solution 𝑥ℎ
generated during the evolution is evaluated by the following weighed average:
𝑢
∑𝑘𝑗=1 𝑑(𝑥ℎ , 𝑥𝑗 ) 𝑓(𝑥𝑗 ) (2.2)
𝑓(𝑥ℎ ) = 𝑢
∑𝑘𝑗=1 𝑑(𝑥ℎ , 𝑥𝑗 )

where 𝑑(𝑥ℎ , 𝑥𝑗 ) is the Euclidean distance between 𝑥ℎ and 𝑥𝑗 , 𝑘 is the number of neighbor
solutions, and 𝑢 = 2.

The number of neighbors k and the size of the database D significantly impact the
computational costs of the algorithm, since they determine the number of operations to perform
the similarity comparisons and to calculate the approximations. Section 3.2 will present some
considerations regarding the database management.

2.3 Constraint-Handling Techniques

A general constrained optimization problem may be formally defined in a q-dimensional
search space as follows:
minimize 𝑓(𝒙)

subject to 𝑔𝑖 (𝒙) ≤ 0, 𝑖 = 1, … , 𝑚 (2.3)

ℎ𝑗 (𝒙) = 0, 𝑗 = 1, … , 𝑝

𝑙𝑘 ≤ 𝑥𝑘 ≤ 𝑢𝑘 , 𝑘 = 1, … , 𝑞
where 𝒙 = (𝑥1 , 𝑥2 , 𝑥3 , … , 𝑥𝑞 ) is a vector of design variables whose components 𝑥𝑖 present lower
and upper bounds [𝑙𝑘 , 𝑢𝑘 ]. The goal is to minimize the objective function 𝑓(𝒙), considering
inequality and equality constraints (respectively 𝑔𝑖 (𝒙) ≤ 0 and ℎ𝑗 (𝒙) = 0) that define the
feasible region.

Many classical CHTs transform this constrained problem into an unconstrained one. For
this purpose, whenever any given constraint is violated a penalty functions 𝑝(𝒙) is added to the
objective function 𝑓(𝒙); thus, the objective function and constraints are combined into a single
value. Another approach is to maintain those values apart along the evolution. Deb [51] proposed
a binary tournament selection method that ranks the solutions according to the values of the
objective function and constraint violation. That method comprises a pairwise comparison where
a feasible solution is always preferred over an infeasible one; if both solutions are feasible, the
one having the best objective function value is preferred; and if both solutions are infeasible, the
one having the lowest constraint violation value is preferred. Later, other methods have followed
this approach, such as the Stochastic Ranking (SR) method [52], the GCR (Global Competitive
Ranking) method [53], and the Ho and Shimizu ranking method (HSR) [54].
 −6−

2.4 Surrogate-assisted methods for constrained problems

Surrogate-assisted evolutionary methods to solve constrained problems have been

presented in the literature, including the ASRES – Approximated Stochastic Ranking Evolution
Strategy, which employs two levels of approximation along the evolution, to estimate the values
of the objective and penalty functions [55].

The SADE-kNN method presented in [38] employed the binary tournament selection
technique proposed by Deb [51]. Other methods recently presented in the literature also
employed the DE algorithm associated with some of the CHTs described above, including for
instance SA-DECV [21] and SA-DECV_SR [56]. These works presented methods that couple
DE with the k-NN surrogate model to approximate both the values of the objective function and
the sum of constraint violations. The former also employs Deb’s feasibility rules of [51]; the
latter work [56], after comparing different CHTs associated to the same surrogate model,
concludes that the best performance is provided by the Stochastic Ranking (SR) method [52].

Recently, [57] presented the e-DE method following two different approaches for the
crossover of solutions (with the goal of balancing population diversity and fast convergence).
This method employs seven comparison criteria for feasible and unfeasible solutions, comprising
an extension of Deb’s feasibility rules.

3 A new surrogate-assisted differential evolution method with multiple

constraint ranking
This section describes the main characteristic of the new surrogate-assisted evolutionary
method proposed in this work: SADE-kNN-MCR. Those characteristics are related to the
following aspects:

➢ The Multiple Constraint Ranking (MCR) technique, comprising a specialized CHT to deal
with the large number of complex constraints that may be found in real-world engineering
problems;

➢ A merit-based strategy to manage the database that stores the exact solutions employed by
the surrogate model;

➢ A strategy that triggers a random reinitialization or “rebirth” of the population whenever it

stagnates on local optima.

3.1 The Multiple Constraint-Handling technique (MCR)

Recently, focusing on the solution of complex real-world engineering optimization
problems, we had presented the Multiple Constraint Ranking technique (MCR) in association
with a Genetic Algorithm [39]. Now, in this paper the MCR is associated with the DE algorithm,
comprising the SADE-kNN-MCR method and thus replacing the pair-wise tournament method
of [51] that had been employed in the SADE-𝒌NN method described in Section 2.4 . In MCR,
the fitness function F for each solution 𝑥𝑖 is evaluated as follows:
 −7−

𝑚
𝑗
𝑅𝑁𝑣𝑖 + ∑ 𝑅𝜙𝑖 , if only infeasible solutions
𝑗=1
𝐹(𝑥𝑖 ) = 𝑚
𝑗 (3.1)
𝑅𝑓𝑖 + 𝑅𝑁𝑣𝑖 + ∑ 𝑅𝜙𝑖 , otherwise
{ 𝑗=1

where rank Rf sorts the value of the objective function f, and RNv sorts the number of constraints
violated. MCR extends the rank-based approach described in Section 2.3: while previous
ranking methods such as the HSR [54] employ only one rank R associated to all violation
values (built as a single summation of the values for functions 𝑣𝑗 (𝑥) of all constraints), here
𝑗
MCR splits R into several ranks 𝑅𝜙 , one for each constraint 𝑗 = 1, … , 𝑚, building multiple
queues considering the objective function and the violation of each constraint. Algorithm 1
presents the pseudocode of the MCR.

The HSR and other previous rank-based techniques may face difficulties in complex
problems characterized by several constraints with different units and/or different orders of
magnitude; in those cases, the single rank of constraint violation values could be dominated by a
given constraint with higher violation values. Of course, this would not be an issue for simpler
cases where one knows a-priori the ranges of minimum and maximum violation values, and
usual normalization procedures could be applied. However, for many real-world engineering
applications such information is not available and cannot be estimated in advance [58,59,1,2,60].
With MCR, all types of constraints are given the same priority and importance in the evaluation
of each solution, irrespective of the range of their violation values.

Algorithm 1: Pseudocode of the MCR CHT

Initialize population of 𝑥𝑖 individuals (𝑝𝑜𝑝𝑠𝑖𝑧𝑒);
for 𝑖 = 1, 𝑝𝑜𝑝𝑠𝑖𝑧𝑒 do
Evaluate objective function and constraints of each individual 𝑥𝑖 ;
end
for each solution 𝑥𝑖 do
𝑅𝑁𝑣𝑖 = ranking position of 𝑥𝑖 according to its number of violations;
for each constraint 𝑗 do
𝑗
𝑅𝜙𝑖 += position of 𝑥𝑖 in rank according to violation in constraint 𝑗;
end
if There are feasible individuals in the population then
𝑅𝑓𝑖 = ranking position of 𝑥𝑖 according to its objective function;
𝑗
𝐹(𝑥𝑖 ) = 𝑅𝑓𝑖 + 𝑅𝑁𝑣𝑖 + 𝑅𝜙𝑖 ;
else
𝑗
𝐹(𝑥𝑖 ) = 𝑅𝑁𝑣𝑖 + 𝑅𝜙𝑖 ;
end
end

The MCR presents other advantages: it does not require any user-defined parameter; also,
when there are no feasible solutions in the population, F depends only on the ranks associated to
the constraints ( RNv and Rj ) , thus the computational costs associated to evaluations of the
objective function f is avoided.
 −8−

3.2 Surrogate model: Database management

A difficult and nontrivial task associated to the k-NN metamodel described in section 2.2 is
the definition of an adequate database size to store the exact solutions. Extensive databases may
slow down the evolution; many similar individuals can be generated, compromising the
population diversity. On the other hand, small databases may not efficiently approximate the
different solutions scattered through the search space. This requires an efficient database
management considering the strategies for insertion, maintenance, and replacement of the
solutions to keep the database size constant.

Usually, an insertion step choses the best solution of the population for evaluation by the
exact model [61-63]. In this work, a merit-based database management strategy is employed to
define how the solutions are replaced according to their contribution to the approximation,
following some guidelines that have been suggested in [64]. Initially, a predetermined number of
solutions are generated and evaluated by the exact function to compose the database D. Those
with the best objective function values also compose the initial population (here it is assumed
that the number of solutions in the population is less or equal than the number of solutions in the
database). Evolution occurs considering the current population evaluated by the surrogate model,
and the database solutions used in the approximation process are scored as indicated in
Algorithm 2 (that shows the general structure of the metamodel including this merit-based
strategy). After each generation, a percentage of solutions of the current population is randomly
chosen; they are evaluated by the exact function and introduced into the database, replacing the
lower scored solutions.

Algorithm 2: Pseudocode of the 𝑘-NN metamodel with the merit-based strategy

for 𝑖 = 1, 𝑝𝑜𝑝𝑠𝑖𝑧𝑒 do
Evaluate the similarity considering the solutions of database 𝐷;
Select the most similar 𝑘 solutions;
Define the updated fitness function according to the arithmetic average of the 𝑘 distance-weighted
neighboring solutions
Increase with +1 the score of the selected solutions of 𝐷;
Decrease with -1 the score of the not selected solutions of 𝐷;
end
Evaluate a percentage of the population by the exact function;
Replace the lower scored solutions in D by the solutions evaluated by the exact function;

3.3 Avoiding stagnation around local optima

A very common problem associated to optimization algorithms in general is the stagnation

of solutions around a local minimum in the search space. For a wide class of problems, the
distribution of the solutions in the search space may be a decisive factor for this stagnation
behavior, when many evaluations of the objective function may be uselessly performed.
An example of this behavior may be observed in Table 1 that presents results of the
application of the SADE-kNN-MCR to the G1 function (from the G-Suite benchmark functions
proposed for the IEEE-CEC 2006 competition on real parameter constrained optimization [42]).
 −9−

The success rate (SuccRate) of the algorithm is 0.84; that is, it finds the optimal solution -
15.0000 in 21 out of the 25 runs. In those 21 successful runs, the algorithm performs an average
of 3,783 evaluations of the objective function (succ_nfe). This value corresponds to
approximately 0.76% of the maximum number of evaluations (500,000) that defines the stopping
criterion of the algorithm. In the other 4 runs, the algorithm remains stagnated in the local
optimum -12.0000 until it reaches the termination criterion of 500,000 evaluations. This way, the
average number of evaluations considering all 25 runs (nfe) increases to 83,177.22. This does not
accurately represent the performance of the algorithm: although it was able to reach the optimum
with a small number of evaluations in most of the runs (21), the stagnation in the other 4 runs
drastically increases the value of this metric (nfe).

Table 1. Results for the G1 function

Func Best Average Worst succ_nfe succRate Nfe

G1 -15.0000 -14.5999 -12.0000 3,783 0.84 83,177.22

There are several strategies to avoid this problem; one of them consists in restarting the
population along the evolution, generating new candidate solutions whenever the population
converges to a local optimum. The strategy employed in this work maintains in the database the
population from the previous generation, and triggers the reinitialization (or “rebirth”) of the
population after a given number of generations in which stagnation of the best solution is
observed. In the context of the SADE-kNN-MCR algorithm, this strategy may be particularly
suited to avoid the frequent premature convergence behavior due to the following reasons: The
new population may visit new regions of the search space, counteracting the tendency of being
stuck in regions where it previously stagnated; and there is no additional computational cost,
since the objective function of the new individuals from the restarted population are
approximated by the k-NN metamodel using the values from the solutions stored in the k-NN
database, and thus there is no need for costly extra evaluations of the objective function.

To summarize, the method proposed in this work (that combines the DE algorithm with a
k-NN surrogate model and the MCR constraint-handling technique) is referred to as SADE-
kNN-MCR. Its variation that includes the population reinitialization described in this section is
called SADE-kNN-MCR+.
 − 10 −

4 Numerical experiments
The performance of the proposed methods is compared with the other methods recently
presented in the literature that have been mentioned in Section 2.4: ASRES, SADE-kNN, SA-
DECV, SA-DECV_SR and e-DE. For this purpose, two sets of benchmark numerical
experiments are executed: the twenty-four G-Suite benchmark functions proposed for the IEEE-
CEC 2006 competition on real parameter constrained optimization [42]; and a suite of five
benchmark mechanical engineering problems [65]: Pressure vessel [66], Welded beam [67],
Cantilever beam [68], Speed reducer and Spring design [65]. The main parameters related to
these latter problems are presented in Table 2, where 𝑛 is the number of decision variables, LI is
the number of linear inequality constraints, NI is the number of nonlinear inequality constraints.
LE and NE are the number of linear and nonlinear equality constraints, respectively. The
complete formulation and mathematical definition of these benchmark problems are already well
documented in those references; thus, they will not be reproduced here.

Table 2. Parameters of the engineering problems

Type of
Problem n LI NI LE NE
function
Pressure Vessel [66] 4 Quadratic 3 1 0 0
Welded Beam [67] 4 Quadratic 6 1 0 0
Tension/Compression Spring [65] 3 Quadratic 1 3 0 0
Cantilever Beam [68] 10 Quadratic 6 5 0 0
Speed Reducer [65] 7 Cubic 1 10 0 0

To assess specifically the advantages of the MCR constraint-handling technique described

in Section 3.1, special attention will be focused on the problems that are representative of many
actual engineering problems, i.e. presenting constraints with different units and/or different
orders of magnitude: functions G10 and G21 from the CEC2006 suite, the Pressure Vessel and
the Welded Beam engineering problems. Lower and upper bounds for the constraint functions of
most of those problems can be analytically calculated, as shown in Table 3; the differences in
scale and magnitude are evident. The orders of magnitude have been numerically estimated for
the Welded Beam problem that presents severely nonlinear constraint functions.

Table 3. Lower and upper bounds for the constraints with different magnitudes
G10 G21
-0.95 ≤ 𝑔1 ≤ 4 -1,000 ≤ 𝑔1 ≤ 640.227
-3.45 ≤ 𝑔2 ≤ 3.975 -48,250 ≤ ℎ1 ≤ 52,500
-10.9 ≤ 𝑔3 ≤ 8.9 -47,625 ≤ ℎ2 ≤ 39,448
-10,065,000 ≤ 𝑔4 ≤ 1,748,999.187 -0.303071 ≤ ℎ3 ≤ 0.384611
-11,227,500 ≤ 𝑔5 ≤ 11,227,500 -0.4085 ≤ ℎ4 ≤ 0.497
-11,240,000 ≤ 𝑔6 ≤ 11,215,000 -1.6449 ≤ ℎ5 ≤ 1.7146
Pressure Vessel Welded Beam (Orders of magnitude)
-3.85375 ≤ 𝑔1 ≤ 4.9807 −104 ≤ 𝑔1 ≤ 104
4
-1.90175 ≤ 𝑔2 ≤ 4.9046 −10 ≤ 𝑔2 ≤ 106
-1,288,673.3 ≤ 𝑔3 ≤ 57,317,600 −10 ≤ 𝑔3 ≤ 10
40 ≤ 𝑔4 ≤ 230 −10 ≤ 𝑔4 ≤ 102
−10 ≤ 𝑔5 ≤ 10−1
−1
−10 ≤ 𝑔6 ≤ 104
7
−10 ≤ 𝑔7 ≤ 103
 − 11 −

The remainder of this section will present the main parameters considered for the
execution of the methods for those benchmark problems and will describe the approaches for the
assessment of their performance.

4.1 Parameters of the algorithms

Following the suggestion of [42], the algorithms are tested from the results of twenty-five
independent executions performed for each benchmark problem. Also following the
specifications of [42], the tolerances for satisfying the equality constraints, and for determining if
the optimal solution x* has been found, are respectively 0.0001 and f(x*) – f(x) < 0.0001. The
maximum number of function evaluations is set to 500,000 for the G-Suite problems. Regarding
the structural engineering problems, 10,000 function evaluations are considered to obtain the
results presented in section 5.2 to compare the different SADE-kNN algorithms.

Numerical experiments have been conducted to set the value for the parameter that control
the reinitialization of the population, i.e. the number of iterations in which stagnation is observed
(as described in section 3.2). The results indicated that an appropriate value would be 1,000
iterations. Similar experiments have also indicated that an appropriate value for the percentage of
solutions selected in the current population at each generation, for the evaluation of their
objective function and introduction in the database (as described in section 3.2), is twenty-two
percent. Table 4 presents the values for the remaining parameters employed for the SADE-kNN
algorithm, and also the parameters that have been used by the other algorithms being compared.

Table 4. Parameters of the algorithms

SA-DECV SA-DECV-SR ASRES e-DE
SADE-kNN
[21] [56] [55] [57]
Pop. size (N) 30 90 90 200 50
Differential Runs 25 30 25 25 25
Evolution CR 0.9 1 1 - 0.9
F 0.8 0.9 0.9 - 0.8
NN
k (number of n (problem n (problem
neighbors)
2 regression -
dimension) dimension)
k-NN model
D (database
size)
1.5*N 2*N 2*N N -

4.2 Performance Assessment

The performance of the methods is assessed by direct comparisons and by nonparametric
statistical tests, as described next.

4.2.1 Direct Comparisons

Direct comparisons are performed by observing the most relevant performance measures
collected in tables: best, average, worst results for 25 runs and average of the number of function
evaluations to find a feasible solution, considering only the successful runs (succ_nfe); ratio
between the number runs in which the algorithms find the optimum solution and the total number
 − 12 −

of runs (succRate); average of the number of function evaluations (nfe) (when succRate = 1, then
succ_nfe = nfe); ratio of runs where the algorithm finds a feasible solution (feasRate).

4.2.2 Sign Test

The Sign Test (ST) is a statistical procedure that indicates if the results of two methods are
statistically different. It employs pairwise comparisons that count the number of cases on which
a method wins. The method that provides a better result for a given problem receives a positive
sign, while the other receives a negative sign. The significance of the difference between these
pairwise comparisons (or significance level) is tested by calculating probability values (p-values)
[69] from the totalized signs. If the p-value is higher than a given threshold value α, then the null
hypothesis (𝐻0 ) that both methods are equivalent is accepted; otherwise 𝐻0 is rejected, and the
performance of the methods is different. The most commonly used threshold values are 0.01 and
0.05 [70-73], representing a significance level of respectively 1% and 5%, i.e., the chance of
rejecting the null hypothesis when in fact it is correct.

4.2.3 Performance Profiles

The comparisons also use another nonparametric statistical test, specifically the
Performance Profiles (PPs) that were introduced in [74] to evaluate and compare the
performance of a set of S solvers for a given set of optimization problems P. The PPs are a useful
tool for the easy visualization and interpretation of results from computational experiments.
Briefly speaking, they comprise graphs with several curves (one for each solver s  S
considering all problems p  P), built considering a given metric mp,s . This metric can be the
number of evaluations of the objective function; or a statistical parameter such as mean or best
value computed from objective function values obtained in several independent runs of the
method. The performance ratio rp,s, is then calculated dividing this metric mp,s by the
corresponding value of the metric obtained by the solver that performed best on problem p. An
overall assessment of the performance of solver s may then be given by:
size  p  P : rp ,s    (4.1)
 s ( ) =
np
where np is the total number of problems in set P; and  s ( ) ∈ →[0, 1] may be seen as the
cumulative distribution function for the performance ratio rp,s; that is, the probability of solver s
to have its ratio rp,s within a factor τ of the best possible ratio.
For =1,  s (1) indicates the accuracy of solver s, being a normalized measure of the
number of times it was able to provide the best value for the considered metric, or the probability
that it will win over all other solvers in set S;  s (1) = 1 indicates that this solver wins over all
other solvers in all problems P.
For larger values of ,  s ( ) indicates the robustness of the solver, i.e., the number of
times it is able to provide feasible solutions; considering an upper limit rM as an arbitrarily large
value assigned to rp,s when solver s is not able to find a solution for problem p, then  s (rM ) = 1
and the probability that method s solves a problem is given by s* = lim s ( ) .
 →rM
 − 13 −

Thus, the PPs are curves [  s ( ) ,], nondecreasing, piecewise constant functions,
continuous from the right at each discontinuity point. A detailed explanation including an
illustrative example of the assembly of a PP may be found in [39].

5 Comparing the SADE-kNN methods

Initially, this section compares the proposed methods (SADE-kNN-MCR and SADE-
kNN-MCR+) with the SADE-kNN method previously presented by the authors in [38]. The
performance is compared by their application to the sets of numerical experiments listed at the
beginning of Section 4: the G-Suite benchmark functions proposed for the IEEE-CEC 2006
competition and the suite of five benchmark mechanical engineering problems.

5.1 G-Suite functions

5.1.1 Direct Comparisons

Table 5 presents the direct comparisons between the methods, in terms of the performance
measures defined in Section 4.2.1, statistically evaluated from the results of the 25 independent
runs. Those measures are: best, average and worst solution; succ_nfe: average number of
function evaluations considering only the successful runs; succRate: ratio between successful
and total runs; nfe: average number of evaluations considering all 25 runs; feasRate: ratio of
feasible runs and total runs. Results for functions G20 and G22, for which no algorithm has
currently found a feasible solution, are not presented. Values in bold indicate the results in which
SADE-kNN-MCR+ finds better or equal solutions than the other algorithms. For problems in
which all three methods reached at least one successful run, the cells corresponding to succ_nfe
shows in parenthesis the percentage values corresponding to the reduction/increase of the
succ_nfe values provided by SADE-kNN-MCR+ in relation to the smaller value observed for
the other methods.

Table 5. Results obtained by the SADE-kNN algorithms for the G-Suite problems

Function SADE-kNN SADE-kNN

SADE-kNN
[best known] MCR MCR+
best -15 -15 -15
average -14.5599 -14.5999 -15
worst -12 -12 -15
G1
succ_nfe 3,722.4 3,783 2,528.5 (-32%)
[-15.0000]
succRate 0.84 0.84 1
nfe 83,127.32 83,177.22 2,528.5
feasRate 1 1 1
best -0.7429 -0.7727 -0.8036
average -0.6367 -0.7458 -0.7729
worst -0.5277 -0.6532 -0.7186
G2
succ_nfe - - 28,632.7
[-0.8036]
succRate 0 0 0.16
nfe 500,000 500,000 424,581.2
feasRate 1 1 1
 − 14 −

Function SADE-kNN SADE-kNN

SADE-kNN
[best known] MCR MCR+
best -0.4515 -0.1687 -1.0005
average -0.0399 -0.0288 -0.4379
worst 0.0000 0.0000 -0.0487
G3
succ_nfe - - 169,089.4
[-1.0005]
succRate 0 0 0.24
nfe 500,000 500,000 380,873.3
feasRate 1 0.68 1
best -30,665.5386 -30,665.5386 -30,665.5386
average -30,665.5386 -30,665.5386 -30,665.5386
worst -30,665.5386 -30,665.5386 -30,665.5386
G4
succ_nfe 2,597.60 2,457.12 1,208.2 (-51%)
[-30,665.5386]
succRate 1 1 1
nfe 2,597.60 2,457.12 1,208.2
feasRate 1 1 1
best 5,126.49 5,126.49 5,126.49
average 5,126.49 5,165.92 5,126.49
worst 5,126.4977 6,112.22 5,126.49
G5
succ_nfe 17,809.7 26,003.9 14,612.6 (-18%)
[5,126.49]
succRate 0.88 0.96 1
nfe 74,855.24 44,963.92 14,612.6
feasRate 0.96 1 1
best -6,961.8138 -6,961.8138 -6,961.8138
average -6,961.8138 -6,961.8138 -6,961.8138
worst -6,961.8138 -6,961.8138 -6,961.8138
G6
succ_nfe 1,234.88 1,134.76 655.5 (-42%)
[-6,961.8138]
succRate 1 1 1
nfe 1,234.88 1,134.76 655.5
feasRate 1 1 1
best 24.3073 24.3082 24.3062
average 28.2727 48.2589 24.3062
worst 121.7574 512.2665 24.3062
G7
succ_nfe - - 64,067.2
[24.3062]
succRate 0 0 1
nfe 500,000 500,000 64,067.2
feasRate 1 1 1
best -0.09582 -0.09582 -0.09582
average -0.09582 -0.09582 -0.09582
worst -0.09582 -0.09582 -0.09582
G8
succ_nfe 291.76 326.48 182.8 (-37%)
[-0.09582]
succRate 1 1 1
nfe 291.76 326.48 182.8
feasRate 1 1 1
best 680.638 680.6353 680.630
average 681.287 680.6783 680.630
worst 682.476 680.7954 680.630
G9
succ_nfe - - 38226.8
[680.630]
succRate 0 0 1
nfe 500,000 500,000 38226.8
feasRate 1 1 1
best 7,049.249 7,049.249 7,049.2480
G10
average 7,278.785 7,198.919 7,049.2480
[7,049.2480]
worst 11,100.000 8,332.225 7,049.2480
 − 15 −

Function SADE-kNN SADE-kNN

SADE-kNN
[best known] MCR MCR+
succ_nfe - - 20,198.4
succRate 0 0 1
nfe 500,000 500,000 20,198.4
feasRate 1 0.96 1
best 0.74990 0.74990 0.74990
average 0.9695 0.7736 0.74990
worst 1.0000 1.0000 0.74990
G11
succ_nfe 2,995.0 2,941.57 1,334.2 (-55%)
[0.7499]
succRate 0.12 0.56 1
nfe 440,363.28 221,649.32 1,334.2
feasRate 1 0.76 1
best -1 -1 -1
average -1 -1 -1
worst -1 -1 -1
G12
succ_nfe 385.96 439.52 211.5 (-45%)
[-1.0000]
succRate 1 1 1
nfe 385.96 439.52 211.5
feasRate 1 1 1
best 0.05394 0.05394 0.05394
average 0.3534 0.2930 0.05394
worst 1.000 1.000 0.05394
G13
succ_nfe 43,906.85 55,912.83 48,682.1 (+10%)
[0.05394]
succRate 0.28 0.48 1
nfe 372,297.24 286,840.32 48,682.1
feasRate 1 0.96 1
best -47.76 -47.76 -47.76
average -47.595 -47.513 -47.76
worst -43.845 -43.844 -47.76
G14
succ_nfe 55,179.31 100,064.78 54,356.1 (-1%)
[-47.76]
succRate 0.88 0.56 1
nfe 108,558.52 276,038.16 54,356.1
feasRate 1 1 1
best 961.7150 961.7150 961.7150
average 961.7150 961.7150 961.7150
worst 961.7150 961.7150 961.7150
G15
succ_nfe 1,1431.3 1,7245.28 5,412.4 (-53%)
[961.7150]
succRate 0.92 1 1
nfe 54,909.56 17,245.28 5,412.4
feasRate 0.92 1 1
best -1.9051 -1.9051 -1.9051
average -1.9051 -1.9051 -1.9051
worst -1.9051 -1.9051 -1.9051
G16
succ_nfe 4,633.36 4,330.66 1,447.8 (-67%)
[-1.9051]
succRate 1 0.96 1
nfe 4,633.36 2,4157.6 1,447.8
feasRate 1 0.96 1
best 8,853.53 8,853.53 8,853.53
average 8,868.96 8,910.50 8,860.07
G17 worst 8,927.59 9,011.91 8,938.51
[8,853.53] succ_nfe 69,887.31 70,850.44 258,603.70 (+73%)
succRate 0.76 0.40 0.68
nfe 173,115.36 345,508.28 304584
 − 16 −

Function SADE-kNN SADE-kNN

SADE-kNN
[best known] MCR MCR+
feasRate 0.96 0.96 0.84
best -0.866 -0.866 -0.866
average -0.8654 -0.8649 -0.866
worst -0.8536 -0.8597 -0.866
G18
succ_nfe 253,743.70 85,907.04 5,855.1 (-93%)
[-0.8660]
succRate 0.40 0.92 1
nfe 401,500.28 119,034.84 5,855.1
feasRate 1 1 1
best 32.6632 32.659 32.6569
average 39.2811 38.080 32.6586
worst 81.044 70.502 32.6676
G19
succ_nfe - - -
[32.6555]
succRate 0 0 0
nfe 500,000 500,000 500,000
feasRate 1 1 1
best 193.7546 193.7546 193.7245
average 193.7546 193.7546 199.4284
worst 193.7546 193.7546 324.7028
G21
succ_nfe - - 122454
[193.7245]
succRate 0 0 0.6
nfe 500,000 500,000 258,371.4
feasRate 0.64 0.92 0.92
best -400.055 -400.055 -400.055
average -400.055 -154.191 -400.055
worst -400.055 -0.0024 -400.055
G23
succ_nfe 68,852.00 88,843.33 51,922.6 (-25%)
[-400.055]
succRate 0.04 0.24 1
nfe 482,757.88 401,325.76 51,922.6
feasRate 0.04 0.96 1
best -5.508 -5.508 -5.508
Mean -5.5080 -5.5080 -5.508
worst -5.5080 -5.5080 -5.508
G24
succ_nfe 765.20 710.56 370 (-48%)
[-5.5080]
succRate 1 1 1
nfe 765.20 710.56 370
feasRate 1 1 1

It can be seen in Table 5 that the SADE-kNN-MCR+ method was able to find the optimal
solution for 21 out of the 22 functions. Although it had not reached the known optimum for
problem G19, the error is relatively small. Also, the fact that the best and worst solutions are
very close (that is, the results found in the twenty-five runs present a small variation) indicate the
robustness of the method. Considering these 21 functions for which the SADE-kNN-MCR+ was
able to reach the optimal solution, it frequently (except for function G17) obtained the highest
success rate (succRate), indicating its best performance in terms of the greater number of runs in
which it can find the optimum.
The other methods were able to reach the global optimum for only 15 out of these 21
functions (failing in functions G2, G3, G7, G10, G19 and G21). Considering those 15 functions
for which all methods found the optimal solution, the highest succRate (always equal to 1) is
 − 17 −

again obtained by the SADE-kNN-MCR+. The feasibility rate (feasRate) obtained by the
SADE-kNN-MCR+ is again always equal to 1.

Considering the 6 problems for which only SADE-kNN-MCR+ was able to find the
optimal solution, in three of them (G7, G9 and G10), it obtained a success rate of 1 (that is, all
runs reached the optimal solution).

Regarding the G10 and G21 problems that present constraints with different units and/or
different orders of magnitude, shown in Table 3: for function G10, the enhanced performance of
the SADE-kNN-MCR method can be measured by the better average and worst metrics
compared with the original SADE-kNN; then, the association of the “rebirth” strategy makes the
SADE-kNN-MCR+ obtain the optimal solution in all runs. For function G21, the enhanced
performance of the SADE-kNN-MCR and SADE-kNN-MCR+ methods is reflected by their
higher number of feasible runs (FeasRate).

Besides the increased performance in finding optimal solutions as commented above, the
SADE-kNN-MCR+ also reduces the average number of function evaluations to find a feasible
solution (succ_nfe). To assess this behavior, in the cells corresponding to succ_nfe for SADE-
kNN-MCR+ and considering only the 15 test problems for which all three methods reached at
least one successful run, Table 5 shows percentage values in parenthesis that correspond to the
ratio between the succ_nfe values provided by SADE-kNN-MCR+, and the smaller succ_nfe
value observed for the other methods. The number of evaluations is reduced for 13 out of those
15 problems. For the other two problems (G13 and G17), although requiring a higher number of
evaluations, SADE-kNN-MCR+ presents better success rates; for the G13 function, for instance,
it finds the optimal solution in 100% of the runs, while SADE-kNN succeeds in only 28%.

5.1.2 Sign Test

Table 6 summarizes the SuccRate values for all methods, considering the 15 test problems
for which all three methods reached at least one successful run. For all those problems (except
G17) SADE-kNN-MCR+ found the optimal solution for 100% of the runs, while the other
methods reached SuccRate = 1 for only 6 out of those 15 problems. The results of the Sign Test
from the pairwise comparison between SADE-kNN-MCR+ and each other method are shown in
the last three lines of Table 6 (discarding the draws). The number of wins and losses confirms
the observations from Table 5, i.e., the SADE-kNN-MCR+ outperforms all other methods. All
p-values (computed according to the procedure described in Section 4.2.2) are small enough to
reject the null hypothesis, meaning the pairwise compared samples are not statistically equivalent
in all cases. This demonstrates that SADE-kNN-MCR+ outperforms all other methods.
 − 18 −

Table 6. Success Rate

SADE- SADE-kNN-
Problem SADE-kNN-MCR+
kNN MCR
1 G1 0.84 0.84 1
2 G4 1 1 1
3 G5 0.88 0.96 1
4 G6 1 1 1
5 G8 1 1 1
6 G11 0.12 0.56 1
7 G12 1 1 1
8 G13 0.28 0.48 1
9 G14 0.88 0.56 1
10 G15 0.92 1 1
11 G16 1 0.96 1
12 G17 0.76 0.40 0.68
13 G18 0.40 0.92 1
14 G23 0.04 0.24 1
15 G24 1 1 1
SADE-kNN-MCR+ Wins (+) 8 9
SADE-kNN-MCR+ Losses (-) 1 0
p-values 0.019531 0.001953

5.1.3 Performance Profiles

The SuccRate values presented in Table 6 were also used as the metric for the generation
of the performance profiles, according to the procedure described in Section 4.2.3 . The PPs
shown in Figure 1 again indicate that SADE-kNN-MCR+ is the best performer, since its curve
reaches 𝜌(𝜏) = 1 with the lowest 𝜏 value. A global indication of the performance of the methods
may also be observed in Figure 2 that contains a bar chart where the y axis indicates the areas
under the PP curves, again confirming the better accuracy and robustness of SADE-kNN-
MCR+.

Figure 1. Performance Profiles – SADE-kNN methods

 − 19 −

Figure 2. Global performance – SADE-kNN methods

5.2 Suite of structural engineering problems

Table 7 compares the results of the methods for the suite of five benchmark mechanical
engineering problems shown in Table 2. All algorithms provided feasible solutions in all runs.
Values in boldface indicate the best results amongst the algorithms.

Table 7. Results for the suite of engineering problems – SADE-kNN methods

SADE-kNN-MCR+ SADE-kNN-MCR SADE-kNN

Best 6,059.7143 6,059.7143 6,059.7180
Pressure Vessel Average 6,176.5486 6,577.5020 6,276.0322
Worst 6,820.4100 7,332.8415 7,332.8529
Best 1.724852 1.724852 1.726247
Welded Beam Average 1.724852 1.802360 1.784099
Worst 1.724852 2.497832 2.319204
Best 64,578.194 64,578.194 65,036.880
Cantilever Average 66,581.615 66,815.056 70,160.740
Worst 68,597.130 68,597.130 74,223.488
Best 2,996.3481 2,996.3481 2,996.3481
Speed Reducer Average 2,996.3481 3,002.1815 2,999.8637
Worst 2,996.3481 73,035.625 3,035.6255
Best 0.012665 0.012665 0.012675
Tension/Compr. Average 0.013285 0.013095 0.012952
Spring
Worst 0.025651 0.017773 0.015754

Regarding the “best” solutions, both SADE-kNN-MCR+ and SADE-kNN-MCR are the
best performers, except for the Speed Reducer problem where all of them achieved the same
solution. It is important to stress that, for the Pressure Vessel and Welded Beam problems that
present constraints with different units and/or different orders of magnitude, the better
 − 20 −

performance of the MCR-based methods compared with the SADE-kNN algorithm may be due
to the increased efficiency of the MCR constraint-handling method in dealing with this type of
problems, as indicated in [39]. The increased robustness of the SADE-kNN-MCR+ compared
with the other methods is demonstrated by the average and worst results, since it presents the
best performance for all problems except the Tension/Compression Spring.

Here, since the number of examples (only five benchmark problems) is low, statistical
performance assessment methods such as the sign test cannot be adequately employed. Even so,
it may be illustrative to compare the performance profiles for the methods. Since SADE-kNN-
MCR+ and SADE-kNN-MCR found the same best values for all problems, the average was
employed as the comparison metric to be evaluated by the performance profiles. Figure 3 shows
that SADE-kNN-MCR+ finds best values for 80% of the problems with 𝜌(1) = 0.8. Altogether,
SADE-kNN-MCR+ is the best overall performer, as also indicated by Figure 4.

Figure 3. Performance Profiles – SADE-kNN methods

applied to the engineering problems. Metric: average.

Figure 4. Global performance – SADE-kNN methods

applied to the engineering problems. Metric: average.
 − 21 −

6 Comparison with other recent methods

Since in the previous section SADE-kNN-MCR+ had demonstrated its better performance
amongst the proposed SADE methods, this section now compares its results with those obtained
by other methods recently presented in the literature. Those methods have been mentioned in
Section 2.4: ASRES, SA-DECV, SA-DECV_SR and e-DE which employ different surrogate
models and/or different constraint handling techniques. The performance of the methods is
compared firstly by their application to the IEEE-CEC 2006 G-Suite benchmark functions, and
then to the benchmark of mechanical engineering problems as described in the beginning of
Section 4.

6.1 G-Suite functions

6.1.1 Direct comparisons

Table 8 compares the results of the methods, in terms of the available main statistical
parameters that characterize their performance: the best solution found amongst the runs; the
average number of function evaluations considering the successful runs succ_nfe; and the ratio
between successful and total runs SuccRate. The cells corresponding to succ_nfe shows also, in
parenthesis, the percentage values corresponding to the reduction/increase of the succ_nfe values
provided by SADE-kNN-MCR+ in relation to the smaller value observed for the other methods.

Table 8. Comparison of Results

SA-
SADE-kNN- SA-DECV ASRES e-DE
DECV_SR
MCR+ [21] [55] [57]
[56]
Best -15 -15 -15 -15 -15
G1 2,528.5
succ_nfe 7,972 15,730.00 35,406 64,274
[-15.0000] (-68%)
SuccRate 1 0.83 0.84 1 1
Best -0.8036 -0.713 -0.6252 -0.739 − 0.8036
G2 28,632.7
succ _nfe - - - 192,297
[-0.8036] (-85%)
SuccRate 0.16 0 0 0 0.92
Best -1.0005 -0.333 -0.3127 -0.998 − 1.0005
G3 169,089.4
succ _nfe - - - 33,066
[-1.0005] (+80%)
SuccRate 0,24 0 0 0 1
Best -30,665.5386 -30,665.539 -30,665.5387 -30,665.539 − 30,665.5386
G4 1,208.2
succ _nfe 5,113 7,175 15,104 43,942
[-30,665.5386] (-76%)
SuccRate 1 1 0,8 1 1
Best 5,126.49 5,126.49 5,126.49 5,126.49 5,126.49
G5 14,612.6
succ _nfe 69,418 21,960 19,281 152,110
[5,126.49] (-24%)
SuccRate 1 1 0.96 1 1
G6 Best -6,961.8138 -6,961.814 -6,961.8138 -6,961.814 − 6,961.8138
[-6,961.8138] succ _nfe 655.5 7,335 5,734 9,603 42,098
 − 22 −

SA-
SADE-kNN- SA-DECV ASRES e-DE
DECV_SR
MCR+ [21] [55] [57]
[56]
(-86%)
SuccRate 1 1 1 1 1
Best 24.3062 24.409 24.3062 24.306 24.3062
G7 64,067.2
succ _nfe - 29,440 77,674 99,614
[24.3062] (+54%)
SuccRate 1 0 0.04 0.08 1
Best -0.09582 -0.096 -0.09582 -0.096 − 0.09582
G8 182.8
succ _nfe 676 539.3 1,027 6,254
[-0.09582] (-66%)
SuccRate 1 1 0.96 1 1
Best 680.630 680.630 680.630 680.63 680.630
G9 38,226.8
succ _nfe 9,831 7,994 30,618 29,446
[680.630] (+79%)
SuccRate 1 0.2 0.84 1 1
Best 7,049.2480 7,260.041 7,049.2480 7,049.408 7,049.2480
G10 20,198.4
succ _nfe - 45,560 - 135,934
[7,049.2480] (-56%)
SuccRate 1 0 0.04 0 1
Best 0.7499 0.75000 0.7499 0.75 0.7499
G11
succ _nfe 1,334.2 10,478 10,580 2,792 24,566
[0.7499]
SuccRate 1 1 0.8 1 1
Best -1 -1 -1 -1 -1
G12 211.5
succ _nfe 1,139 660.6 2,996 1,354
[-1.0000] (-68%)
SuccRate 1 1 0.64 1 1
Best 0.05394 0.453 0.337683 0.054 0.05394
G13 48,682.1
succ _nfe - - 11,292 303,741
[0.05394] (+77%)
SuccRate 1 0 0 0.84 0.44
Best -47.76 -47.24 -47.48 -47.76 − 47.76
G14 54,356.1
succ _nfe - - 92,820 92,730
[-47.76] (-41%)
SuccRate 1 0 0 0.08 1
Best 961.715 961.715 961.715 961.715 961.715
G15 5,412.4
succ _nfe 46,105 12,430 8,519 95,022
[961.7150] (-36%)
SuccRate 1 0.63 0.64 1 1
Best -1.905 -1.905 -1.905 -1.905 − 1.905
G16 1,447.8
succ _nfe 10,149 9,384 16,179 24,410
[-1.9051] (-85%)
SuccRate 1 1 1 1 1
Best 8,853.53 8,872.62 8,857.65 8,853.54 8,853.53
G17 258,603.71
succ _nfe - - 21,490 264,232
[8,853.53] (+92%)
SuccRate 0.68 0 0 0.76 0.44
Best -0.8660 -0.8660 -0.8660 -0.8660 − 0.8660
G18 5,855.1
succ _nfe 32,751 29,830 40,840 140,458
[-0.8660] (-80%)
SuccRate 1 0.03 0.76 0.92 1
 − 23 −

SA-
SADE-kNN- SA-DECV ASRES e-DE
DECV_SR
MCR+ [21] [55] [57]
[56]
Best 32.6569 35.404 32.6556 32.665 32.6555
G19
succ _nfe - - 53,740 - 207,286
[32.6555]
SuccRate 0 0 0.04 0 1
Best 193.7245 193.732 193.7267 - 193.7245
G21 122,454
succ _nfe - - - 84,500
[193.7245] (+31%)
SuccRate 0.6 0 0 0 0.72
Best -400.055 -308.803 -365.2985 -348.816 -400.055
G23 51,922.6
succ _nfe - - - 46,922
[-400.055] (+10%)
SuccRate 1 0 0 0 1
Best -5.508 -5.508 -5.508 -5.508 − 5.508
G24
succ _nfe 370 (-88%) 3,444 3,015 3,638 19,678
[-5.5080]
SuccRate 1 1 1 1 1

In Table 8, it is interesting to compare the performance between the SA-DECV and the
SA-DECV-SR methods. The latter is a variation that uses a rank-based constraint handling
technique, instead of simple feasibility rules; thus, unlike the SA-DECV, the SA-DECV-SR
provides optimal solutions to problems G7, G10 and G19. It also improves the solution found by
SA-DECV in problem G21. SA-DECV-SR also produced a slight improvement in the number
of optimal executions.

However, the performance of these algorithms is still inferior to those presented by SADE-
kNN-MCR+ and e-DE: SA-DECV methods failed in functions G13, G14, G17 and G23, while
SADE-kNN-MCR+ and e-DE found optimal solution for most problems. Moreover, they found
good solutions in most executions, as indicated by their SuccRate.

Considering the problems that present constraints with different orders of magnitude, such
as G10 and G21 problems, SADE-kNN-MCR+ presents an enhanced performance. For function
G10, besides finding the optimal solution in all runs, it also requires the smaller number of
function evaluations.

Table 9 summarizes the main results of Table 8, collecting the number of functions for
which each method was able to find the optimal solutions in at least one run, and required the
lower number of function evaluations to find the optimal solution.

Table 9. Summary of some performance measures

SADE-kNN- SA- SA-DECV-

N. of functions where: ASRES e-DE
MCR+ DECV SR
Found optimal solution
21 12 15 16 22
in at least one run
Required lower number
14 0 2 2 4
of function evaluations
 − 24 −

A more comprehensive comparison of the performance of all methods will be presented in

terms of nonparametric statistical tests previously described in section 4.2: the sign test, and the
performance profiles.

6.1.2 Sign Test

Table 10 presents the results of the Sign Test based on the pairwise comparison between
SADE-kNN-MCR+ and each other method in terms of the average number of function
evaluations (succ_nfe). The number of wins and losses confirms the observations based on Table
8, i.e., the SADE-kNN-MCR+ outperforms all other methods. The statistical assessment in
terms of the p-values indicates that all of them are below the threshold, which confirms that the
superiority of the SADE-kNN-MCR+ is statistically significant.

Table 10. Sign Test, Pairwise Comparisons

SADE-kNN-MCR+ versus SA-DECV SA-DECV-SR ASRES e-DE

# of wins / # of losses 20 / 1 19 / 3 18 / 3 17 / 5
(p-value) (0.00001) (0.0004) (0.0007) (0.0084)

Regarding the success rate (succRate), both SADE-kNN-MCR+ and e-DE reach 100% of
optimal executions in 16 problems. e-DE gets more optimal runs in 4 out of the remaining 6.
However, this leads to a p-value of 0.3437, not small enough to indicate that there is a
statistically significant difference between these two methods regarding this particular metric.

6.1.3 Performance Profiles

The succ_nfe values were also used as the metric for the generation of the performance
profiles shown in Figure 5, according to the procedure described in Section 4.2.3 . A global
indication of the performance of the methods may also be observed in the bar chart of Figure 6
where the y axis indicates the areas under the PP curves.

Figure 5. Performance Profiles

 − 25 −

Figure 6. Global performance of the methods

In Figure 5 it is interesting to observe that, for smaller values of , the curve

corrresponding to the e-DE method is located below those for the SA-DECV-SR and ASRES
algorithms; however, since e-DE is able to find feasible solutions for more problems than SA-
DECV, SA-DECV-SR and ASRES, it ends up with the second larger area in the bar chart of
Figure 6. The overall best performer is SADE-kNN-MCR+, since its curve approaches 𝜌(𝜏) = 1
with the lowest 𝜏 value and presents the larger area in the bar chart of Figure 6, reflecting the fact
that it requires fewer evaluations for the greater number of problems.

6.2 Suite of structural engineering problems

Finally, this section compares the results for some of the benchmark engineering problems
shown in Table 2. Here the proposed SADE-kNN-MCR+ method is compared only with SA-
DECV-SR, since results for the other methods could not be found in the literature. Anyway, this
is a useful comparison since both methods share the following characteristics: 1) DE is used as
the search algorithm; 2) Some of the solutions are approximated by a similarity-based surrogate
model; and 3) Rank-based constraint handling techniques are employed.

For these comparisons, the maximum number of function evaluations is set to 30,000
function evaluations for the Welded Beam problem; 20,000 for the Pressure Vessel; and 5,000
for the Tension/Compression Spring problem.

Table 11 compares the statistical analysis of the results for the three benchmark problems
considering 25 runs: the best, average and worst solutions amongst all runs; and the average
number of function evaluations considering the successful runs. Unlike section 5.2, where
optimal reference values were not used as a stopping criterion, the results presented in
parentheses were considered as optimal in order to compare the average number of function
evaluations that each algorithm required to reach it.
 − 26 −

Table 11. Results for the suite of engineering problems

Prob. Algorithm Best Average Worst Succ_nfe

SADE-kNN-
Welded Beam MCR+
1.7248523 1.7248523 1.7248523 3,983.0
(1.7248523) SA-DECV-SR 1.72486 1.72496 1.72595 6,971
SADE-kNN-
Pressure Vessel MCR+
6,059.71436 6,062.20310 6,090.52622 6,719.7
(6,059.71433) SA-DECV-SR 6,059.7 6,296.0336 6,820.44 6,139
Tension/Compression SADE-kNN-
MCR+
0.012732 0.012765 0.012902 1,409.4
Spring
(0.01266523) SA-DECV-SR 0.0127127 0.013047148 0.0143943 1,617

For the Welded Beam problem, SADE-kNN-MCR+ provided the better solution as
indicated in the “best” column of Table 11. It also obtained this optimal solution in all runs, as
indicated by the worst value, which is equal to the best one. Also, its average number of function
evaluations (3,983.0) is nearly half of the value required by the SA-DECV-SR.

Considering the Pressure Vessel problem, SADE-kNN-MCR+ also provided better

solutions regarding all statistical measures. It obtained the accuracy required for the optimal
solution in 20 out of the 25 runs, employing an average number of 6,719.7 function evaluations.
In [56] it is not clear whether the value of 6,139 evaluations refers to the number of function
evaluations considering only the optimal runs; anyway, it can be concluded that SADE-kNN-
MCR+ is the most efficient algorithm in obtaining the best solutions considering all runs.

Finally, for the Tension/Compression Spring problem, SADE-kNN-MCR+ provided

better results regarding all measures (average, worst, and number of function evaluations),
except for the best solution. The best solution provided by SADE-kNN-MCR+ (0.012732)
reached the first termination criterion (f(x*) – f(x) < 0.0001), thus stopping the evolution process.

Table 12 presents the values of the variables corresponding to the optimal solutions
provided by SADE-kNN-MCR+ . Again, the number of problems is low, thus the sign test
cannot be adequately employed. However, it may be illustrative to compare the performance
profiles for the SADE-kNN-MCR+ and SA-DECV_SR algorithms. Figures 8 and 9 show
respectively the individual and global performance using as metric the succ_nfe values presented
in Table 11. Those results show that SADE-kNN-MCR+ is the best performer, since not only it
gives better results for two problems (𝜌(1) = 0.666), but still finds optimal executions in all of
them (there is 𝜏 such that 𝜌(𝜏) = 1).

Table 12. Values of the variables corresponding to the optimal solutions

provided by SADE-kNN-MCR+

Welded Beam Pressure Vessel Tens./Compr. Spring

H 0.205730 𝑇𝑠 0.8125 d 0.050016
L 3.470489 𝑇ℎ 0.4375 D 0.317587
T 9.036624 R 42.098446 N 14.025337
B 0.205730 L 176.636597
f(x*) 1.724852 f(x*) = 6,059.714361 f(x*) = 0.012732
 − 27 −

Figure 7. PPs comparing Figure 8. Global performance of

SADE-kNN-MCR+ and SA-DECV-SR SADE-kNN-MCR+ and SA-DECV-SR

7 Conclusions and extensions

This work presented a new merit-based surrogate-assisted evolutionary method designed
for applications to complex real-world constrained optimization problems. Besides presenting a
very good accuracy in finding optimal solutions, the main goal of the method is to reduce the
number of objective function evaluations.

Those goals of accuracy and computational cost savings are pursued by the following
approaches: a) employing the Multiple Constraint Ranking (MCR) technique to deal with the
large number of constraints with different orders of magnitude and/or different units that are
common in the aforementioned complex engineering problems; b) devising an enhanced
approximation model based on the 𝑘-nearest-neighbor (𝒌-NN) algorithm associated to a merit-
based database management procedure; and c) implementing a strategy that triggers a random
reinitialization of the population whenever it stagnates on local optima. These approaches have
the potential to present significant cost savings by reducing the number of objective function
evaluations: the MCR does not require evaluations when there is no feasible solution in the
population, and the merit scheme of the k-NN model evaluates only part of the population.

Two variants of the proposed method (SADE-kNN-MCR and SADE-kNN-MCR+) were

applied to a set of 24 constrained benchmark problems and 5 mechanical engineering problems.
Based on direct comparisons of different performance measures; nonparametric statistical sign
tests; and performance profile graphs, the results were compared with those provided by other
algorithms recently presented in the literature that also employ surrogate models and constraint
handling techniques (SADE-kNN, ASRES, SA-DECV, SA-DECV_SR and e-DE). The
comparisons have shown that both variants of the proposed method presented superior accuracy
and drastically reduced the number of function evaluations. The best performer is SADE-kNN-
MCR+ since in several executions it was able to escape from local optima.

This method may be particularly suited for real-life engineering problems such as, for
instance, the design of offshore systems that have been addressed in [58,59,1-3]. Those are
 − 28 −

complex engineering problems defined on high-dimensional spaces, with many design variables,
nonlinear objective and constraint functions (the latter comprised by constraints with different
orders of magnitude and/or different units), and requiring evaluations by expensive Finite
Element nonlinear dynamic analyses; thus, reducing the number of evaluations is a critical issue.
The potential of the SADE-kNN-MCR+ method to solve such practical engineering problems is
indicated by its performance in the Welded Beam and Pressure Vessel problems, which are also
characterized by severely nonlinear constraint functions with diverse magnitudes. Thus,
extensions of this work will incorporate this method into optimal design procedures for mooring
systems, submarine pipelines and risers that have been considered in [1,3,60]. It is expected that
the resulting procedures will lead to increased efficiency in terms of accuracy, computational
costs, and more importantly to better engineering design solutions.

Declarations

Funding

The authors acknowledge the support of the Brazilian funding agencies CNPq (grant
numbers 306069/2016, 306186-2017, 308873/2017-3); FAPERJ (grant numbers E-
26/200.314/2016, E-26/202.767/2017); and CAPES (Coordenação de Aperfeiçoamento de
Pessoal de Nível Superior), finance code 001.

Conflicts of interests

The authors declare that they have no conflict of interest.

Contributions

The authors contributed to each part of this paper equally. The authors read and approved the
final manuscript.

Ethical approval

This paper does not contain any studies with human participants or animals performed by any of
the authors.

References
1. Vieira IN, de Lima BSLP, Jacob BP (2012) Bio-inspired algorithms for the optimization of offshore oil
production systems. International Journal for Numerical Methods in Engineering 91 (10):1023-
1044. doi:https://doi.org/10.1002/nme.4301

2. de Lucena RR, Baioco JS, de Lima BSLP, Albrecht CH, Jacob BP (2014) Optimal design of submarine
pipeline routes by genetic algorithm with different constraint handling techniques. Advances in
Engineering Software 76:110-124. doi:https://doi.org/10.1016/j.advengsoft.2014.06.003

3. Monteiro BdF, de Pina AA, Baioco JS, Albrecht CH, de Lima BSLP, Jacob BP (2016) Toward a
methodology for the optimal design of mooring systems for floating offshore platforms using
evolutionary algorithms. Marine Systems & Ocean Technology 11:55-67.
doi:https://doi.org/10.1007/s40868-016-0017-8
 − 29 −

4. Queipo NV, Haftka RT, Shyy W, Goel T, Vaidyanathan R, Tucker PK (2005) Surrogate-based analysis
and optimization. Paper presented at the Progress in aerospace sciences,

5. Saunders C, Gammerman A, Vovk V (1998) Ridge regression learning algorithm in dual variables.

6. Hendrickx W, Gorissen D, Dhaene T Grid enabled sequential design and adaptive metamodeling. In:
Simulation Conference, 2006. WSC 06. Proceedings of the Winter, 2006. IEEE, pp 872-881

7. Kybic J, Blu T, Unser M (2002) Generalized sampling: a variational approach. I. Theory. IEEE
Transactions on Signal Processing 50:1965-1976

8. Kybic J, Blu T, Unser M (2002) Generalized sampling: a variational approach. II. Applications. IEEE
Transactions on Signal Processing 8:1977-1985

9. Mullur AA, Messac A (2006) Metamodeling using extended radial basis functions: a comparative
approach. Engineering with Computers:203

10. Lowe D (1988) Multi-variable functional interpolation and adaptive networks. Complex Systems:321-
355

11. Myers RH, Montgomery DC, Anderson-Cook CM (2016) Response surface methodology: process
and product optimization using designed experiments. John Wiley & Sons,

12. Ferrari S, Stengel RF (2005) Smooth function approximation using neural networks. IEEE
Transactions on Neural Networks 16:24-38

13. Emmerich MT, Giannakoglou KC, Naujoks B (2006) Single-and multiobjective evolutionary
optimization assisted by gaussian random field metamodels. IEEE Transactions on Evolutionary
Computation 10

14. Ulmer H, Streichert F, Zell A Evolution strategies assisted by Gaussian processes with improved
preselection criterion. In: Evolutionary Computation, 2003. CEC'03. The 2003 Congress on, 2003.
IEEE, pp 692-699

15. Simpson TW, Mauery TM, Korte JJ, Mistree F (2001) Kriging models for global approximation in
simulation-based multidisciplinary design optimization. AIAA journal 39:2233-2241

16. Kecman V (2005) Support vector machines--an introduction. Support vector machines: theory and
applications:605-605

17. Krempser E, Bernardino HS, Barbosa HJ, Lemonge AC Differential evolution assisted by surrogate
models for structural optimization problems. In: Proceedings of the international conference on
computational structures technology (CST), 2012.

18. Krempser E, Bernardino HS, Barbosa HJ, Lemonge AC (2017) Performance evaluation of local
surrogate models in differential evolution-based optimum design of truss structures. Engineering
Computations 2 (34):499-547

19. Fonseca LG, Barbosa HJC, Lemonge ACC (2009) A similarity-based surrogate model for enhanced
performance in genetic algorithms. Opsearch 46 (1):89-107

20. Liu Y, Sun F (2011) A fast differential evolution algorithm using k-Nearest Neighbour predictor.
Expert Systems with Applications 38(4):4254-4258

21. Miranda-Varela M-E, Mezura-Montes E (2016) Surrogate-assisted Differential Evolution with an

Adaptive Evolution Control Based on Feasibility to Solve Constrained Optimization Problems.
Paper presented at the Fifth International Conference on Soft Computing for Problem Solving,
 − 30 −

22. Grefenstette JJ, Fitzpatrick JM Genetic search with approximate fitness evaluations. In: Proc. Of the
Intl. Conf. on Genetic Algorithms and Their Applications, 1985. pp 112-120

23. Ratle A Optimal sampling strategies for learning a fitness model. In: Proceedings of the 1999
Congress on Evolutionary Computation-CEC99 (Cat. No. 99TH8406), 1999. IEEE, pp 2078-2085

24. Hong YS, Lee H, Tahk MJ (2003) Acceleration of the convergence speed of evolutionary algorithms
using multi-layer neural networks. Engineering Optimization 35 (1):91-102

25. Sun C, Jin Y, Zeng J, Yu Y (2014) A two-layer surrogate-assisted particle swarm optimization
algorithm. Soft Computing 19 (6):1461-1475. doi:10.1007/s00500-014-1283-z

26. Anderson KS, Hsu Y Genetic crossover strategy using an approximation concept. In: In Proceedings
of the 1999 Congress on Evolutionary Computation-CEC99 (Cat. No. 99TH8406), 1999. IEEE, pp
527-533

27. Rasheed K, Vattam S (2002) Comparison of methods for using reduced models to speed up design
optimization. Paper presented at the 4th Annual Conference on Genetic and Evolutionary
Computation,

28. de Pina AC, Albrecht CH, de Lima BSLP, Jacob BP (2014) Wavelet network meta-models for the
analysis of slender offshore structures. Engineering Structures 68:71-84.
doi:https://doi.org/10.1016/j.engstruct.2014.02.039

29. de Pina AC, Monteiro BdF, Albrecht CH, de Lima BSLP, Jacob BP (2014) ANN and wavelet
network meta-models for the coupled analysis of floating production systems. Applied Ocean
Research 48:21--32. doi:https://doi.org/10.1016/j.apor.2014.07.009

30. Storn R, Price K (1997) Differential Evolution – A Simple and Efficient Heuristic for Global
Optimization over Continuous Spaces. Journal of Global Optimization 11:341-359

31. Wang YS, Shi YJ, Yue BX, Teng HF An efficient differential evolution algorithm with approximate
fitness functions using neural networks. In: International Conference on Artificial Intelligence and
Computational Intelligence, Heidelberg, Berlin., 2010. Springer, pp 334-341

32. Pahner U, Hameyer K Adaptive coupling of differential evolution and multiquadrics approximation
for the tuning of the optimization process. In: IEEE Transactions on magnetics, 2000. pp 1047-
1051.

33. Mallipeddi R, Lee M (2015) An evolving surrogate model-based differential evolution algorithm.
Applied Soft Computing 34:770-787. doi:10.1016/j.asoc.2015.06.010

34. Jin C, Qin AK, Tang K (2015) Local ensemble surrogate assisted crowding differential evolution. In
2015 IEEE Congress on Evolutionary Computation (CEC). IEEE,

35. Liu B, Koziel S, Zhang Q (2016) A multi-fidelity surrogate-model-assisted evolutionary algorithm for
computationally expensive optimization problems. Journal of computational science, vol 12.

36. Awad NH, Ali MZ, Mallipeddi R, Suganthan PN (2018) An improved differential evolution algorithm
using efficient adapted surrogate model for numerical optimization. Information Sciences 451-
452:326-347. doi:10.1016/j.ins.2018.04.024

37. Yang Z, Qiu H, Gao L, Jiang C, Zhang J (2019) Two-layer adaptive surrogate-assisted evolutionary
algorithm for high-dimensional computationally expensive problems. Journal of Global
Optimization 74 (2):327-359. doi:10.1007/s10898-019-00759-0

38. Garcia RdP, de Lima BSLP, Lemonge ACdC (2017) A Surrogate Assisted Differential Evolution to
Solve Constrained Optimization Problems. Paper presented at the IEEE Latin American
Conference on Computational Intelligence (LA-CCI), Arequipa, Peru,
 − 31 −

39. Garcia RdP, de Lima BSLP, Lemonge ACdC, Jacob BP (2017) A rank-based constraint handling
technique for engineering design optimization problems solved by genetic algorithms. Computers
and Structures 187:77-87. doi:https://doi.org/10.1016/j.compstruc.2017.03.023

40. Lampinen J, Zelinka I On stagnation of the differential evolution algorithm. In: Proceedings of
MENDEL, 2000. pp 76-83

41. Hrstka O, Kučerová A (2004) Improvements of real coded genetic algorithms based on differential
operators preventing premature convergence. Advances in Engineering Software 35(3-4):237-246

42. Liang JJ, Runarsson TP, Mezura-Montes E, Clerc M, Suganthan PN, Coello CAC, Deb K (2006)
Problem Definitions and Evaluation Criteria for the CEC 2006 Special Session on Constrained
Real-Parameter Optimization. Journal of Applied Mechanics 41(8):8-31

43. Storn R System design by constraint adaptation and differential evolution. In: IEEE Transactions on
Evolutionary Computation, 1999. pp 22-34

44. Bergamaschi PR, Saramago SDFP, dos Santos Coelho L (2008) Comparative study of SQP and
metaheuristics for robotic manipulator design. Applied numerical mathematics 58(9):1396-1412

45. Das S, Suganthan PN (2011) Differential evolution: A survey of the state-of-the-art. Paper presented
at the IEEE transactions on evolutionary computation,

46. Wu G, al. e (2018) Ensemble of differential evolution variants. Information Sciences:172-186

47. Kenneth V (1999) Price, An introduction to differential evolution, New ideas in optimization.
McGraw-Hill Ltd., UK, Maidenhead, UK,

48. Price K, Storn R, M, Lampinen JA (2006) Differential evolution: a practical approach to global
optimization. Springer Science \& Business Media,

49. Shepard D A two-dimensional interpolation function for irregularly-spaced data. In: ACM (ed)
Proceedings of the 1968 23rd ACM national conference, 1968. pp 517-524

50. Hu H, Lee DL (78-91) Range nearest-neighbor query. IEEE Transactions on Knowledge and Data
Engineering 18

51. Deb K (2000) An efficient constraint handling method for genetic algorithms. Computer Methods in
Applied Mechanics and Engineering 186:311-338

52. Runarsson T, Yao X (2000) Stochastic ranking for constrained evolutionary optimization. IEEE Trans
Evol Comput 4:284–294

53. Runarsson TP, Yao X Continuous selection and self-adaptive evolution strategies. In: Proceedings of
the 2002 Congress on Evolutionary Computation - CEC'02, 2002. pp 279-284

54. Ho PY, Shimizu K (2007) Evolutionary constrained optimization using an addition of ranking method
and a percentage-based tolerance value adjustment scheme. Information Sciences 177 (14):2985-
3004. doi:10.1016/j.ins.2007.01.011

55. Runarsson TP (2006) Approximate Evolution Strategy using Stochastic Ranking. Paper presented at
the 2006 IEEE International Conference on Evolutionary Computation, Vancouver, BC, Canada,
July 16-21

56. Miranda-Varela M-E, Mezura-Montes E (2018) Constraint-handling techniques in surrogate-assisted

evolutionary optimization. An empirical study. Applied Soft Computing 73:215-229.
doi:10.1016/j.asoc.2018.08.016
 − 32 −

57. Yu X, Lu Y, Wang X, Luo X, Cai M (2019) An effective improved differential evolution algorithm to
solve constrained optimization problems. Soft Computing 23 (7):2409-2427.
doi:https://doi.org/10.1007/s00500-017-2936-5

58. de Lima BSLP, Jacob BP, Ebecken NFF (2005) A hybrid fuzzy/genetic algorithm for the design of
offshore oil production risers. International Journal for Numerical Methods in Engineering 64
(11):1459-1482. doi:https://doi.org/10.1002/nme.1416

59. de Pina AA, Albrecht CH, de Lima BSLP, Jacob BP (2011) Tailoring the particle swarm optimization
algorithm for the design of offshore oil production risers. Optimization and Engineering 12 (1-
2):215-235. doi:https://doi.org/10.1007/s11081-009-9103-5

60. Baioco JS, de Lima Jr. MHA, Albrecht CH, de Lima BSLP, Jacob BP, Rocha DM (2018) Optimal
Design of Submarine Pipelines by a Genetic Algorithm with Embedded On-Bottom Stability
Criteria. Mathematical Problems in Engineering 2018:1-21.
doi:https://doi.org/10.1155/2018/1781758

61. Regis RG (2014) Evolutionary programming for high-dimensional constrained expensive black-box
optimization using radial basis functions. IEEE Transactions on Evolutionary Computation 18:326-
347

62. Regis RG, Shoemaker CA (2007) A stochastic radial basis function method for the global
optimization of expensive functions. INFORMS Journal on Computing 19:497-509

63. Elsayed SM, Ray T, Sarker RA A surrogate-assisted differential evolution algorithm with dynamic
parameters selection for solving expensive optimization problems. In: Evolutionary Computation
(CEC), 2014 IEEE Congress on, 2014. IEEE, pp 1062-1068

64. da Silva AF (2016) Optimization of constrained problems using Particle Swarm algorithm aided by
surrogate models [in Portuguese]. Doctoral thesis - Federal University of Rio de Janeiro. Advisors:
de Lima BSLP, Lemonge ACdC, Rio de Janeiro

65. Gandomi AH, Yang X-S (2011) Benchmark Problems in Structural Optimization. In: Koziel S, Yang
X-S (eds) Comput. Optimization, Methods and Algorithms. Springer-Verlag, Berlin Heidelberg, pp
259–281

66. Sandgren E (1988) Nonlinear integer and discrete programming in mechanical design. Paper
presented at the ASME Design Technology Conference, Kissimee, FL,

67. Deb K (1991) Optimal design of a welded beam via genetic algorithms. AIAA Journal 29 (11):2013-
2015

68. Erbatur F, Hasançebi O, Tütüncü I, Kılıç H (2000) Optimal design of planar and space structures with
genetic algorithms. Computers & Structures 75(2):209-224

69. Gibbons JD, Chakraborti S (2003) Nonparametric Statistical Inference. Statistics: a Series of
Textbooks and Monographs, Fourth edn. Marcel Dekker, Inc., New York - Basel

70. Fisher RA Theory of statistical estimation. In: Mathematical Proceedings of the Cambridge
Philosophical Society, 1925. Cambridge University Press,

71. Corder GW, Foreman DI (2009) Nonparametric statistics: an introduction. In: Nonparametric
Statistics for Non-Statisticians: A Step-by-Step Approach. John Wiley & Sons, Hoboken, NJ, USA,
pp 101-111.

72. Derrac J, García S, Molina D, Herrera F (2011) A practical tutorial on the use of nonparametric
statistical tests as a methodology for comparing evolutionary and swarm intelligence algorithms.
Swarm and Evolutionary Computation 1 (1):3-18. doi:10.1016/j.swevo.2011.02.002
 − 33 −

73. Lynn N, Suganthan PN (2015) Heterogeneous comprehensive learning particle swarm optimization
with enhanced exploration and exploitation. Swarm and Evolutionary Computation 24:11-24.
doi:10.1016/j.swevo.2015.05.002

74. Dolan ED, Moré JJ (2002) Benchmarking optimization software with performance profiles.
Mathematical Programming 91 (2):201-213. doi:10.1007/s101070100263