Neural Computing and Applications (2022) 34:1079–1098

https://doi.org/10.1007/s00521-021-06447-z (0123456789().,-volV)(0123456789().,-volV)

ORIGINAL ARTICLE

Modeling and experimental validation of dry-type transformers

with multiobjective swarm intelligence-based optimization algorithms
for industrial application
Tugce Demirdelen1 • Burak Esenboga1 • Inayet Ozge Aksu2 • Alican Ozdogan1 • Abdurrahman Yavuzdeger3 •

Fırat Ekinci3 • Mehmet Tümay1

Received: 27 October 2020 / Accepted: 17 August 2021 / Published online: 30 August 2021
 The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2021

Abstract
In recent years, the optimum and efficient design of the transformer core and conductive materials is the most significant
issues to overcome the high-temperature problems. The temperature increases on the transformer materials are directly
related to the energy efficiency of it. The overheating of the core and coils of the transformer reduces the amount of energy
to be obtained from the transformer. However, copper, core, eddy current and other losses can be minimized by obtaining
an optimum design of the transformer for maximum efficiency. Thus, the transformer life and the energy efficiency to be
obtained from the transformer are maximized. The temperature rise and temperature distribution of the windings can be
monitored by computer technology and the transformer can be safely overloaded and the production cost can be mini-
mized. Also, the operating life of the transformers can be further increased by specifying hot spot temperatures on the
transformer coils and core. In this study, 3 kVA and 5 kVA Dyn 11 connected 380/220-V dry-type transformers are
optimized by multiobjective swarm intelligence-based optimization methods. The main contribution of this study is to
prevent the overheating of the transformers by reducing the losses in the transformer core and coils and to reduce the costs
of the transformer. The thermal and electromagnetic analyses of the transformers are realized by ANSYS/Maxwell
software program which utilizes the industry-leading ANSYS/Fluent computational fluid dynamics and finite element
method solvers. Finally, the experimental analyses are realized under the loaded conditions for the transformers. The
experimental results are verified with the simulation results. The optimization, modeling, thermal/electromagnetic analysis
and experimental processes are carried out step by step in this study. The transformer manufacturers will realize the
optimum cost, efficiency and thermal analysis before transformers are manufactured.

Keywords Artificial intelligence  Dry-type transformer  Energy efficiency  Metaheuristic optimizations 

Swarm intelligence

1 Introduction

The transformers have recently become an important

component of the transmission and distribution system of
& Burak Esenboga electrical energy. Since dry-type transformers are protected
[email protected] against fire and explosion, industrial applications tend to
1
turn to non-flammable transformers such as dry-type. It is
Department of Electrical and Electronics Engineering, Adana
observed that the use of dry-type transformers significantly
Alparslan Türkeş Science and Technology University,
Adana, Turkey reduces the risk of fire. Furthermore, dry-type transformers
2 are easier to maintain and have lower operating costs.
Department of Computer Engineering, Adana Alparslan
Türkeş Science and Technology University, Adana, Turkey Many experimental and theoretical studies have been car-
3 ried out on dry-type transformer applications due to all
Department of Energy System Engineering, Adana Alparslan
Türkeş Science and Technology University, Adana, Turkey these advantages. In particular, optimization methods

123
1080 Neural Computing and Applications (2022) 34:1079–1098

based on very different optimization algorithms are carried Reactive power control of the electrical system, which is
out on critical parameters of dry-type transformers such as presented single line diagram based on a real application,
weight, cost and temperature. was realized using GA, PSO and differential evolution
One of the most used software in thermal and electro- (DE) optimization algorithms. It was presented that the DE
magnetic field analysis of transformers is ANSYS/Maxwell strategy is superior to other algorithms in reducing system
program since the error rate between experimental and losses and costs and solving reactive power planning (RPP)
simulation results is at a minimum level. Thermal charac- problem [7]. An optimization method based on the differ-
teristics of dry-type cast resin transformers are examined in ential evolution algorithm has been proposed to overcome
detail. Thermal behavior in dry-type transformers is an the optimum power flow problem of a power system. With
important parameter as it directly affects the efficiency of the proposed approach, transmission losses have been
the transformer. The temperature distribution characteris- reduced to a minimum [8]. The paper presents a novel,
tics of the cast resin transformers are modeled with the 3D evolutionary algorithm-based transformer optimization
finite volume-based CFD model created in ANSYS CFX method which can determine the optimal conductor shape
software, which is a FEM-based software. The experi- for the windings during this examined preliminary design
mental results obtained from 800 kVA transformer and the stage. The FEM-based model and obtained performance
results obtained from the proposed model were compared analysis results are presented to calculate the short circuit
to demonstrate the accuracy of these two different impedance, magnetic field distribution and load losses of
approaches. The accurate and short solution time of pro- the transformer [9]. It is calculated and analyzed that power
posed thermal models is presented in detail [1]. The elec- loss of pulsed power electrical network 400 kV/66 kV/
tromagnetic field and thermal analysis of a three-phase 22 kV transformer. Electromagnetic and thermal analyzes
power transformer were carried out using ANSYS Maxwell were carried out in detail by building a 2D model of the
3D and ANSYS Mechanical software. The power trans- power transformer. The accuracy of the proposed model is
former design ready for production is presented with the presented by comparing the simulation and experimental
obtained performance analysis results [2]. In the event of a results [10]. The thermal and electromagnetic analysis
lightning impulse of a dry-type transformer, the dielectric based on the 3D finite element model was carried out to
strength of the insulation material was analyzed using investigate the effect of ferroresonance on temperature and
ANSYS Maxwell software. Obtained data from perfor- electromagnetic field distribution of a single-phase trans-
mance analysis show how winding configuration affects former. Performance analysis results of the transformer are
voltage distribution between turns. The safety margin presented in comparison with simulation and experimental
across the insulation ratio has been improved by 49% with results [11]. A comprehensive review has been carried out
the proposed winding configuration [3]. The magnetic flux on studies on copper losses, eddy current loss, stray loss
density distribution of 1.25 MVA three-phase power and short circuit forces of transformers by building the
transformer was presented by using ANSYS Maxwell FEM-based model. Also, papers using 2D and 3D FEM
software. Moreover, leakage reactance, losses, energy, analysis in performance analysis of transformers are pre-
radial and axial force were calculated. The proposed pro- sented in detail [12].
totype transformer was manufactured by comparing the To obtain more energy-efficient and lesser losses from
data obtained from simulation results with the experimental 1500 kVA transformer is used optimization algorithms
results [4]. The electromagnetic field and thermal field based on decomposition with Dynamical Resource Allo-
analyzes of the transformer are presented in detail using the cation in the transformer design. Iterative Chaotic Map
finite element method. The mathematical model for ther- with Dynamical Resource Allocation (ICMDRA) has been
mic transfer was presented theoretically. Thanks to the identified as a superior algorithm. 200 kVA transformer is
proposed modeling of the electric field and thermal field, it manufactured with the proposed effective methodology
can be observed values for these fields that cannot be [13]. The thermal modeling of 400 kVA cast resin dry-type
obtained in real applications. [5]. The proposed model transformer was presented using particle swarm optimiza-
based on the 3D finite element method is built to perform tion (PSO) algorithm. This proposed algorithm has been
electromagnetic losses and hot spot analysis of a 2500 kVA observed to be superior to the genetic algorithm [14]. The
dry-type transformer. Furthermore, the experimental and results obtained from the genetic algorithm (GA), particle
simulation results were compared losses at the core, high swarm optimization (PSO) and artificial bee colony (ABC)
voltage windings and low voltage windings. Current den- optimization algorithms used for the cast resin dry-type
sity in foil windings and temperature distribution for the transformer design were compared. According to these
dry-type transformer is presented as a result of simulations results, it has been observed that PSO reaches optimal
studies [6]. solution quickly [15].

123
Neural Computing and Applications (2022) 34:1079–1098 1081

The optimum design of transformer parameters directly proposed for the efficiency optimization of a single-phase
affects energy efficiency. The ant/firefly hybrid algorithm dry-type transformer [22].
is presented with performance analysis results to optimize Dry-type transformers are the key elements in the con-
the core shape in transformer design. The magnetic losses, version, and transmission process of electricity because of
cost, dimensions and weight of the power transformer have their superior energy conversion ratio. However, these
been reduced thanks to the proposed hybrid algorithm. transformers have poor heat dissipation performance due to
Thus, it has been observed that the efficiency of the power have no additional cooling systems, which directly affects
transformer increases by increasing the magnetic perfor- transformer life and results in insulation deterioration and
mance with the hybrid algorithm [16]. The evolutionary thermal aging. Also, these transformers are bigger in size
harmony search algorithm is widely used in recent trans- and weight than oil-type transformers of the same power
former design optimization studies. The modified harmony capacity. This increases the production cost of the trans-
search algorithm is proposed to increase the output quality formers. Therefore, reducing the weight and size of the
of the analysis results. The simulation results of the mul- transformers are significant while achieving good perfor-
tiobjective harmony search algorithm and the proposed mance, cost reduction and increased efficiency. The use of
algorithm are compared in order to prove the superiority of optimization methods for optimum modeling and analysis
the proposed algorithm. It is presented the advantages of a in engineering applications is increasing day by day. In
modified HS using the Ricker map approach in transformer recent years, many optimization studies have been realized
design optimization [17]. The voltage drops, overcurrents on transformer optimization in the literature, and they have
and interruptions occurring in power transformers have proven their effectiveness.
been reduced energy quality. The routine tests of trans- The use of heuristic algorithms whose mathematical
formers should be carried out successfully in order to avoid models are inspired by the behavior of living organisms in
these problems in power transformers. The routine tests, nature has increased in solving engineering problems. HS
which are vital for transformers, were performed with and FA algorithms which are heuristic algorithms are
analyzes based on the firefly algorithm. An analysis soft- proposed recently in the literature and frequently applied in
ware program has been developed for routine transformer different engineering fields. The firefly algorithm used in
testing based on the firefly algorithm. The effectiveness of this study is a population-based metaheuristic algorithm
the proposed program has been tested experimentally. In that models the flashing behavior of fireflies living in
addition, according to simulation results, the firefly algo- tropical regions with each other. Due to its good perfor-
rithm’s superiority over other algorithms is observed [18]. mance and simple implementation, this algorithm has been
Transformers are the most important part of electricity used to find optimum values in many studies in the liter-
transmission and distribution systems. Parameters such as ature. FA has two major advantages over other algorithms:
transformer cost, size and efficiency have been considered automatic subdivisioning and the ability to deal with
in the optimum design of transformers. In the literature, multidimensional problems. Owing to its few parameters to
many studies based on artificial intelligence techniques are adjust, easy to understand, realize, and compute, it can be
carried out to optimize these parameters. Artificial intelli- applied to multiple objectives optimization problems. In
gence techniques, which have been used extensively in the this research, the harmony search algorithm model, which
literature in recent years, have been examined compre- is a recently developed heuristic model, was applied to the
hensively for transformer design [19]. The cost of power optimization problem, in addition to the FA algorithm. HS
transformers in electricity transmission and distribution algorithm is based on the principle of obtaining the best
systems is quite high, so many optimization studies have melody in terms of harmony with the notes played by the
been carried out in the literature to reduce the transformer musicians in an orchestra. This algorithm has been used in
cost. A software based on the ant/firefly hybrid algorithm solving various engineering problems. HS algorithm is
has been developed to optimize the power transformer tank easy to implement. This algorithm is less sensitive to the
sizes. Thanks to the proposed hybrid algorithm, the mate- algorithm parameters. This means that no need to fine-tune
rials used in power transformer tanks have been reduced by these parameters to get good solutions. During the HS
5% to 15%, thus reducing the transformer cost [20]. algorithm operation, it generates a new vector after con-
The different optimization methods were used to mini- sidering all available vectors. This provides the algorithm
mize the size and efficiency of dry-type transformer. The to give better solutions. In this study, it is aimed to reduce
invasive weed optimization (IWO) and firefly algorithm the weight and size of the transformer and gain better
(FA) and particle swarm optimization (PSO) methods were performance by redesigning the dry-type transformer
used to optimize the weight of a dry-type transformer and thanks to proposed optimization algorithms. Therefore, the
its relation to the cost of it [21]. The FA algorithm was design parameters in the transformer are determined using
different optimization methods.

123
1082 Neural Computing and Applications (2022) 34:1079–1098

These optimization methods which are newly introduced o2 T o2 T o2 T

in the literature applied first time in the industry. The major Kx 2
þ Ky 2 þ Kz 2 ¼ Q ð1Þ
ox oy oz
problem of the optimization algorithms is adjusting the
initial values of used parameters during the iteration of the r  ðKrTÞ ¼ Q ð2Þ
algorithm. An advantageous feature of HS algorithm is that where r is the gradient operator and also in Eq. (2) is
no need to adjust these parameters during algorithm oper- shown in vector form.
ation. Also, the structure of HS algorithm is flexible and
easy to apply to problems. HS is applied for the first time in o o o
r ¼ e^x þ e^y þ e^z ð3Þ
weight and efficiency optimization of dry-type transformer. ox oy oz
The success of the HS algorithm has been demonstrated where e^x ,e^y and e^z are expressed the unit vectors directed
according to PSO and FA algorithms used in transformer the x-axes, y-axes and z- axes, respectively, in Eq. (3). T is
optimization in the literature. With the proposed method, the temperature at the point analyzed in C. KðW=m  CÞ
transformer manufacturers will be able to control the parameters in Eq. (1) are thermal conductivities of the dry-
transformer model which will be formed according to the 
type transformer. Q ðW m3 Þ has expressed the internal
electrical and mechanical design of the transformer and heat generation per unit area in the transformer high volt-
take the necessary measures to ensure that this design age winding, low voltage winding and cores. In other
works with the most reliable and optimum state of the words, Q is expressed as loss density. While the losses of
transformer before production. the transformer under no-load conditions are observed in
the core, the losses occurring under load are in low and
high voltage windings [25].
2 Formulation of the proposed transformer
2.2 Electromagnetic side
The detailed mathematical modeling of the proposed
transformers is presented to realize optimization algorithms One of the most critical parameters in transformer design is
[23]. Optimization algorithms have been implemented on electromagnetic modeling and performance analysis of the
the critical parameters of this mathematical model to transformer. Thanks to this analysis, winding losses, core
optimize the efficiency and cost of transformers. The losses, current, voltage and magnetic field curves of dry-
mathematical modeling of dry-type transformers has type transformers can be observed. In this paper, the
included the thermal side, electromagnetic side, and opti- electromagnetic analysis of the dry-type transformer is
mization side sections. carried out by using the ANSYS/Maxwell software. This
software is given information to us about AC electromag-
2.1 Thermal side netic, magnetostatic, electrostatic, eddy current and elec-
trode transient problems by using Maxwell equations.
Iron and copper losses during the operation of dry-type These equations are high-performance interactive solutions
transformers cause the temperature of transformers to to solve electric or magnetic field problems of the electrical
increase. The examining and controlling these temperature machinery in a finite region of space with appropriate
rises in transformers involve the subject of more than one boundary conditions and user-specified initial conditions.
engineering field together. For dry-type, oil-filled and gas- The terms of the equations are used based on the solver
filled types of transformers, different methods such as selected such as electrostatic, magnetostatic, eddy current
electromagnetic theory, electrical circuit theory, heat and magnetic transient. In this study, the transient electric
transfer numerical methods can be evaluated together, and solver is selected for the electromagnetic analysis of the
the transformer thermal model is obtained by using these transformers. This solver evaluates time-varying electric
methods. and magnetic fields. The main structure for the electro-
In this paper, the thermal model of 3 kVA and 5 kVA magnetic solution is a tetrahedron. The assembly of all
dry-type transformers is modeled in ANSYS/Maxwell, and tetrahedral is referred to as the finite element mesh of the
also, hot spot temperature analyzes are performed in detail. model or simply the mesh. Mesh plays important role in the
This software uses the finite element method (FEM) which accuracy of the computed results and thus requires higher
is a numerical solution method used to overcome problems mesh resolution in regions where field fields are of interest
about the flow of fluids, electromagnetic field distribution rapidly. ANSYS/Maxwell works on Maxwell’s equations
and heat transfer. This method is governed by the following that are used for the electromagnetic analysis. The elec-
form of the Poisson equation to analyze the heat transfer of tromagnetic field changes occurring in the conducting
the 3D dry-type transformer in solid form as shown Eq. (1) material at the transformer are mathematically expressed
[24].

123
Neural Computing and Applications (2022) 34:1079–1098 1083

by Maxwell’s equations. The equations used in electro- fireflies moving toward each other depending on their
magnetic field analysis are based on Faraday’s law in attractiveness or in a random direction.
Eq. (4), in Ampere’s law according to Maxwell Eq. (5), In the firefly algorithm, there are three basic rules: (1)
Gauss’s magnetic law in Eq. (6), and Gauss’s electric law All fireflies are considered as unisexual. In other words,
in Eq. (7) [26]. These equations are as follows; gender is not taken into account when calculating the
oB attractiveness and brightness values. (2) The attractiveness
rxE ¼  ð4Þ of a firefly is proportional to its brightness. (3) The
ot
brightness of a firefly or the intensity of light is determined
oD
rxH ¼ J þ ð5Þ by the value of the fitness function. If both fireflies are
ot similar in brightness, the fireflies may move randomly.
rB¼0 ð6Þ The attractiveness of fireflies is defined as:
rD¼q ð7Þ I ðr Þ ¼ I0 egr
2
ð11Þ
oq Here, I and I0 define the light intensity at the specifies
rJ ¼ ð8Þ
ot
where E ðV=mÞ is the electric field intensity, B ðWb=m2 Þ distance and the original light intensity at the source,
is the magnetic flux density, H ðA=mÞ is the magnetic field respectively. r denotes the distance, and g denotes the light
intensity, J ðA=m2 Þ is the conduction current density, absorption coefficient. Similarly, firefly’s light attractive
D ðC=m2 Þ is the electric flux density, and q ðC=m3 Þ is the function b is calculated as follows:
2
source charge density [24]. If the current density is as in bðr Þ ¼ b0 egr ð12Þ
Eq. (8), it provides a continuity condition current density.
The above equations are solved in ANSYS/Maxwell uti- where b0 is the firefly’s light attractiveness at r = 0. The
lizing a magnetic vector potential approach. distance between the two fireflies is calculated using
Euclidean distance. Given that the fireflies i and jare at
2.3 Optimization side positions xi and xj, the distance (rij) is formulated as
follows:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
In this study, three-different optimization methods are u D
  uX  2
proposed for the efficiency and cost-based optimization of rij ¼ xi  xj  ¼ t xi;d  xj;d ð13Þ
transformers. These methods are presented in detail. d¼1
Particle swarm optimization (PSO) is a population-
based heuristic optimization algorithm developed by Here, D is the number of dimensions. The equation of
Eberhart and Kennedy in 1995. It has been inspired by the
position that allows fireflies to generally move toward
social behavior of bird and fish flocks in nature [27].
fireflies with high brightness and attractiveness is as
In the PSO algorithm, every possible solution is called a
follows:
particle. At the beginning of the algorithm, particles begin
2 
randomly distributed in the search space. The motion of the xi ¼ xi þ b0 egr xj  xi þ aei ð14Þ
particles in the search space is based on their velocity and
where a is a scaling parameter, and e is a random number
position values. When the best solution of the ith particle so
far is pibest , the velocity of the ith particle is defined as vi , from Gaussian distribution. The flow diagram of the
the position of the ith particle is determined as xi , the best algorithm is shown in Fig. 1.
solution of all particles so far is gbest ; the speed and posi- Harmony search algorithm (HSA) is an optimization
tion values of the particles are updated as follows: method developed by Geem et al. in 1995 [29]. The HS
vi ðt þ 1Þ ¼ vi ðtÞ þ c1  r1  ðpibest  xi ðtÞÞ þ c2  r2 algorithm aims to obtain the best melody in harmony with
 ðgbest  xi ðtÞÞ ð9Þ the notes played by the musicians in an orchestra. HSA is
generally designed in 5 steps.
xi ðt þ 1Þ ¼ xi ðtÞ þ vi ðt þ 1Þ ð10Þ
Step 1: Define the problem and define the parameters
Here, c1 and c2 are the acceleration coefficients, and t is
the iteration number. r1 and r2 values are random numbers. In this step, the parameters of the algorithm are
The flow diagram of the algorithm is shown in Fig. 1. determined:
The firefly algorithm (FA) is a heuristic optimization
• Harmony Memory Size (HMS)
technique developed by sensing the social behavior of
• Harmony Memory Considering Rate (HMCR)
fireflies [28]. This algorithm is based on the nature of
• Pitching Adjust Rate (PAR)

123
1084 Neural Computing and Applications (2022) 34:1079–1098

Fig. 1 Modeling, optimization and analysis of the proposed transformers with optimum cost and efficiency features gradually

Then, the fitness function of the optimization problem is 2 3

1. Harmony
described. 6 7
6 2. Harmony 7
6 7
Step 2: Initialize the HS Memory (HM) Harmony Memory (HM) ¼ 6
6 .
7
7
6 .. 7
Firstly, the solution set is generated randomly by con- 4 5
sidering HMS. In this step, N harmonics are created and N. Harmony ð15Þ
stored in harmony memory. After that, the fitness values of 2  3
x1;1 x1;2  x1;M  f1
all individuals in the solution set are calculated. 
6 x2;1
6 x2;2  x2;M  f2 7 7
¼6  7
6 .. .. ..  .. 7
4 . . .  . 5
xHMS;1 xHMS;2    xHMS;M  fHMS

123
Neural Computing and Applications (2022) 34:1079–1098 1085

Here, M defines the total decision variable, and f defines most frequently used values in these method parameters are
the fitness function. seen. In line with this information, the initial population is
formed by assigning randomly within the specified mini-
Step 3: Improvise a new solution mum and maximum value range.
In this step, a new solution vector is created, While the C and s values are decided in the optimization
phase of the study, optimization is performed for three
xnew ¼ ½ xnew;1 xnew;2    xnew;M : different situations. These situations are:
The new harmony vector is produced according to the weight
tones in the harmonic memory and according to completely
In this case, the aim is to find suitable C and s values to
random tones. In this step, the component is selected
obtain minimum weight using our transformer model.
among HM members by using HMCR value. The decision
Therefore, our fitness function is equivalent to the weight
variables (xnew;1 , xnew;2 , …,xnew;M ) are decided as follows:
equation of the transformer.
if ðrndðÞ\HMCRÞ
 FF ¼ minðcostðti ÞÞ ð18Þ
xnew;i 2 x1;i ; x2;i ; :::; x1;HMS
where FF is fitness function, ti ith transformer.
else ð16Þ
xnew;i 2 Xi efficiency.
end In this case, the aim is to obtain maximum efficiency
from the transformer. Therefore, the fitness function is
where rndðÞ is a random generated number. Xi is the set of
equivalent to the efficiency equation of the transformer.
the possible range of values for each decision variable. It is
then checked whether tone adjustment is necessary. PAR FF ¼ maxðefficiencyðti ÞÞ ð19Þ
parameter is used for this operation.
if ðrndðÞ\PARÞ by considering weight and efficiency criteria at the same
xnew;i ¼ xnew;i  rndðÞ  bw time.
else ð17Þ
In this case, the aim is to decide on C and s parameters
xnew;i ¼ xnew;i by considering weight and efficiency criteria at the same
end time. In this case, the suitability function is formed by
taking both criteria into account.
where bw is the s an arbitrary distance bandwidth which is
chosen randomly. As it is understood, HMCR and PAR FF ¼ %50  minðcostðti ÞÞ þ %50  max
ð20Þ
parameters affect the performance of the algorithm. These ðefficiencyðti ÞÞ
two parameters help to find general and local optimum
solutions.
Step 4: Update the Harmony Memory 3 Proposed procedures for dry-type
transformers
In this step, a comparison is made between the new
harmony vector and the worst harmony vector in memory.
The parameters of 3 kVA and 5kVA 380/220 V delta-wye
If the fitness value of the xnew is better than the worst
connected dry-type transformers were extracted by their
harmony, the worst harmony vector is excluded from the
mathematical model. These parameters show the perfor-
harmony memory and the new harmony vector is included
mance analysis of transformers based on the cost and
in.
efficiency of the transformer. With the help of the mathe-
Step 5: Check the Termination Criterion matical model, transformers are optimized based on cost
and efficiency characteristics with different optimization
It is checked whether the given termination criterion is
methods. They are designed by using the electromagnetic
fulfilled. If the criterion is not fulfilled, step 3 and step 4 are
analysis program (ANSYS/Maxwell). Then, the electro-
repeated. The flow diagram of the algorithm is shown in
magnetic and thermal analyses of these transformers were
Fig. 1.
carried out with the help of this program. After all these
The initial values of the parameters used in optimization
stages, transformers are manufactured using the optimum
algorithms are assigned in accordance with the structure of
parameters. The magnetic and thermal tests of the trans-
each method. When the literature studies are examined, the
formers were carried out experimentally to validate the

123
1086 Neural Computing and Applications (2022) 34:1079–1098

experimental analysis results with the simulation results. seen that the PSO algorithm runs faster than the other two
Figure 1 presents the modeling, optimization and analysis algorithms.
of the proposed transformers with optimum cost and effi- Speed is an important criterion in the design phase.
ciency features gradually. Also, there are large differences in cost and efficiency
between the results obtained from the HS algorithm, and
the results obtained with the other proposed algorithms. It
4 Simulation and experimental results is seen that the results obtained from the HS algorithm are
more successful than the results obtained from the PSO
4.1 Optimization algorithm.
When the results obtained from the FA and HS algo-
The parameter values C and s are important in the trans- rithms are examined, it is seen that the HS algorithm is
former’s design process. For this analysis, algorithms the faster and more successful. The fitness function change
HS, the FA, and the PSO are used in the 3 kVA and 5 kVA graph during algorithm calculation for 3 kVA is given in
transformer design phases. Through these algorithms, C Figs. 2, 3 and 4.
and s values of the transformers are determined. Calcula- The fitness function change graphs during algorithm
tions are produced according to three specific criteria calculation for 5 kVA transformer are given in Figs. 5, 6
during the decision-making process: and 7.
The design parameters are optimized in HS, FA and
• According to the efficiency
PSO algorithm according to C and s values of the trans-
• According to the weight
former. Because efficiency and weight change of the
• According to both
transformer are performed depending on the core quality
While the C and s values are decided in the optimization factor and the current density of the windings. As a result
phase of the study, optimization is performed for three of the optimization study, the maximum efficiency and the
different situations. These situations are: minimum weight are obtained at different C and s values
during the optimization. The current density and the core
• In the first case, the calculation is made considering the
quality factor to be chosen must be decided by considering
weight criterion. So in this case, the aim is to minimize
the economic criteria of the transformer manufacturer or
weight.
designer.
• In the second case, the efficiency criterion is taken into
In this study, transformers are manufactured, modeled
account. In other words, the aim is to calculate in a way
and analyzed by using optimum efficiency parameters.
to obtain maximum efficiency.
• In the last case, the aim is to make calculations by
4.2 Modeling and electromagnetic analysis
considering weight and efficiency criteria at the same
of transformers
time.
As a result, the weight-based calculation is directly 3 kVA and 5 kVA optimized transformers are modeled
related to the cost parameter. Therefore, the cost and with FEM-based electromagnetic analysis program.
weight-based calculation means the same situation in this Because transformers are electromagnetic devices and this
paper. program can analyze the nonlinear, time-varying motion of
The values obtained at the end of the calculations are electromagnetic components with high accuracy.
detailed in Table 1 for the 3 kVA transformer in Table 2 for Transformers are modeled with optimized parameters to
the 5 kVA transformer. analyze the thermal behavior and performance of them.
When analyzing the results in-depth, it is shown that the The electromagnetic transient analysis of the designed
results obtained from the HS algorithm are generally better transformer is realized. The current, voltage, loss, magnetic
than the two other approaches in terms of cost basis. By field distribution and flux characteristics of the transformer
comparing the results obtained on the multiobjective line are examined in detail thanks to the electromagnetic tran-
according to the efficiency criterion, the result obtained sient analysis results. The design steps for modeling the
from the HS algorithm is shown to be better. transformers in the ANSYS/Maxwell program are exam-
It is seen that the HS algorithm gives the best results on ined in detail as follows statements. Firstly, the trans-
a cost basis and an efficiency basis obtained for 5 kVA formers are modeled as 3D so the core, primary side and
3-phase dry-type transformers. secondary side coils of the transformers are designed by
These results are the values obtained after 200 iterations using Maxwell-3D shown in Fig. 8.
and 20 runs. The swarm size is selected 10 for each During the design process, the transformers’ cores and
method. When the results are examined in general, it is coils are specified in terms of their characteristic features.

123
Neural Computing and Applications (2022) 34:1079–1098 1087

Table 1 Results obtained from

Harmony Search Firefly PSO
the HS, FA and PSO algorithms
in the determination of C and s Weight based Cost (€) 147.2354 152.8183 159.0287
values in 3 kVA transformer
Efficiency 0.9153 0.9152 0.9100
Weight 26.6439 27.6680 28.5864
s 2.1311 2.0593 2.2
C 7.2901 7.5085 7.7887
Elapsed time (s) 0.069429 0.078259 0.039519
Efficiency based Cost (€) 148.2724 153.7512 178.1641
Efficiency 0.9243 0.9233 0.9146
Weight 27.3007 28.2833 32.3739
s 1.7295 1.7 1.755
C 7.2461 7.4649 8.4681
Elapsed time (s) 0.061770 0.078101 0.039760
Multi-objective Cost (€) 147.3843 153.3609 160.9570
Efficiency 0.9173 0.9210 0.9135
Weight 26.7592 28.0695 29.0968
s 2.0424 1.8015 2.0207
C 7.2799 7.4753 7.8351
Elapsed time (s) 0.064972 0.072939 0.044227

Table 2 Results obtained from

Harmony Search Firefly PSO
the HS, FA and PSO algorithms
in the determination of c and s Weight based Cost (€) 200.1684 202.4951 209.0006
values in 5 kVA transformer
design Efficiency 0.9385 0.9384 0.9377
Weight 35.7596 36.1771 37.3253
s 2.028 2.0092 1.9739
C 7.2756 7.3423 7.5303
Elapsed time (s) 0.062263 0.073768 0.036638
Efficiency based Cost (€) 214.7609 227.6361 229.3567
Efficiency 0.9396 0.9386 0.9383
Weight 38.5677 40.8908 41.1872
s 1.7835 1.7021 1.7
C 7.6667 8.0261 8.0756
Elapsed time (s) 0.063874 0.073269 0.041097
Multi-objective Cost (€) 200.6362 208.2347 209.8777
Efficiency 0.9390 0.9383 0.9378
Weight 35.88 37.2362 37.4974
s 1.9907 1.9415 1.9556
C 7.2838 7.5023 7.5533
Elapsed time (s) 0.064826 0.070353 0.043065

The core materials are assigned as iron. The iron properties Also, the magnetic field – magnetic field vector (B-H)
are edited as used in the material characteristics. These curve and magnetic field – loss (B-P) curve of the trans-
characteristics are the relative permeability, bulk conduc- former core are specified to obtain the core loss model
tivity, core loss model, mass density and Poisson’s ratio. accurately. The HV and LV coils of the transformers are

123
1088 Neural Computing and Applications (2022) 34:1079–1098

Fig. 2 Change of fitness function in the weight-based calculation for 3 kVA transformer

Fig. 3 Change of fitness function in the efficiency-based calculation for 3 kVA transformer

selected as copper material. Finally, the air material is geometry can be analyzed using irregular meshes. The
assigned as a region or a boundary because of the dry assembly of all tetrahedral is referred to as the finite ele-
characteristic of the transformer. ment mesh of the model. This method produces approxi-
ANSYS/Maxwell uses the finite element method (FEM) mate solutions. The analysis solution can be repeated by
to solve Maxwell equations. FEM is widely used in elec- increasing the number of elements, changing the element
trical machine applications for electromagnetic and thermal type, changing the mesh production method or manually
analysis. With this method, complex, inhomogeneous and manipulating the mesh. In this study, the length-based
anisotropic materials can be modeled and complex inside selection mesh operation is applied to the

123
Neural Computing and Applications (2022) 34:1079–1098 1089

Fig. 4 Change of fitness function in the multiobjective-based calculation for 3 kVA transformer

Fig. 5 Change of fitness function in the weight-based calculation for 5 kVA transformer

transformers. This mesh operation will limit the edge thousand (1.000.000) particles thanks to mesh operation so
length of all tetrahedrons (or triangles) formed inside a all particles are electromagnetically analyzed. The
selected solid or sheet object. The maximum length of the designed model is analyzed more accurately thanks to the
element is selected 10 mm. This method will provide to mesh operation. The mesh operation on the transformer
control the size or number of elements inside the selected model is given in Fig. 9.
object. The proposed transformer is divided into a hundred

123
1090 Neural Computing and Applications (2022) 34:1079–1098

Fig. 6 Change of fitness function in the efficiency-based calculation for 5 kVA transformer

Fig. 7 Change of fitness function in the multiobjective-based calculation for 5 kVA transformer

After defining the properties of the materials, the After the transformer model is created, the analysis
transformer coils are energized at the full load condition solution time and the type of solution must be formed. The
according to the rated value of the transformers. The solution type was chosen as a magnetic transient since the
energized state of the coils is shown in Fig. 10. analysis was carried out under a certain period at the full
load conditions. Voltage is applied to the primary coils of

123
Neural Computing and Applications (2022) 34:1079–1098 1091

the transformers under load. The changing magnetic field

occurs in the transformer core. The magnetic field distri-
bution of the transformer core is given in Fig. 11.
The modeled transformers are tested under full load
conditions. In this way, the current, voltage and loss
analyses of the transformers are realized. These analysis
characteristics of the transformers are as following
Figs. 12, 13 and 14.
The voltage drops are observed in the primary and
secondary side of coils because the transformers operate at
full load conditions for a while. Since transformers have no
rotating parts, there are no losses such as friction and wind
losses. Therefore, their efficiency is higher than other
electrical machines. However, as with all electrical
Fig. 8 3D FEM model of the proposed transformer
machines, transformers have losses. These losses are two
types of core losses, hysteresis losses and eddy current
losses. These losses give information on the efficiency and
performance of the designed transformer. The designed
transformers are tested electrically and thermally under the
same load conditions.

4.3 Thermal analysis of transformers

Transformers, like all electrical devices and machines,

become hot during their operation. The main cause of this
warming is losses occurring in the device or machine.
Transformer losses that cause an increase in temperature in
windings are generally copper and iron losses. These losses
increase the winding temperature and, however, cause
additional losses due to increased winding resistance. This
Fig. 9 The length-based inside selection mesh operation
reveals that the transformer cooling is also important for
the operating costs of the transformer. Since the coil tem-
perature is related to the losses occurring in the windings
and the life of the winding insulation varies depending on
the operating temperature, the life of the transformer
changes with the losses. Thus, the coil temperature is an
important parameter affecting the losses of transformers,
the actual turn rate and the life of the transformer.
In this study, the transformers are loaded to observe the
temperature distribution on the transformer core and coils.
ANSYS/Icepack program is used for the thermal analysis
of the transformers. This program provides powerful
cooling solutions that utilize the industry-leading ANSYS
Fluent computational fluid dynamics (CFD) solver for
thermal and fluid flow analyses of electrical and electronic
devices.
The designed model created in ANSYS/Maxwell Elec-
tromagnetic is connected with ANSYS/Icepack Fluid
Dynamics in order to determine thermal specifications. The
Fig. 10 Assignment terminals for energizing of primary coils core material is assigned as resistance shown in Fig. 15.
Because the core material acts as resistance toward losses.
The loss type is based on the loss curve of the core material
(B-P curve) given in Fig. 15. To obtain the transformer

123
1092 Neural Computing and Applications (2022) 34:1079–1098

Fig. 11 The magnetic field distribution of the transformers’ core materials (a) 3 kVA transformer (b) 5 kVA transformer

thermal distribution, the thermal condition is assigned as low voltage winding, high voltage winding, and the right
total power to the coils. The power values are applied leg, middle and right leg of the core. This thermocouple
3000-W and 5000-W to the coils. Thermal analysis results measures with an accuracy of ± 2 C. The ambient tem-
are taken at the full load conditions in simulation and perature during experimental studies is 25C and also the
experimental studies. The ambient temperature is selected same ambient conditions apply to the simulation studies.
25 C for the load operation. Fluid material and region are Figure 18 shows the test laboratory under loaded condi-
the air material due to the dry-type characteristic of the tions. Experimental data are obtained by using the trans-
transformers. The analysis setup is constituted by using former test laboratory. The core and coils temperatures of
flow regime, ambient temperature, losses and airflow the transformers are measured at the seven different points.
velocity in x-, y- and z-directions. The velocities values are The cable connections between the measurement devices
1 m/s in all directions. The flow regime is laminar, and the and transformers are shown in Table 3.
maximum iteration number is selected 40 iterations due to 3 kVA and 5 kVA load is applied to the secondary side
the steady-state condition of the transformer load. Fig- of the transformers for 240 and 450 min. The temperature
ures 16 and 17 show the temperature distribution of the 3 distributions in the steady-state conditions at the core, low
kVA and 5 kVA dry-type transformers. voltage winding, and high voltage winding of the trans-
The determination of the winding temperature and the formers are shown, respectively, in Fig. 19. The trans-
hottest point temperature of the transformer before pro- formers have been reached the steady state at 240 min and
duction minimizes the aging of the transformers with the 450 min, respectively. The comparative of experimental
measures to be taken. Therefore, the transformer manu- and simulation temperature measurement of 3 kVA and 5
facturers can control the hottest spots and temperature rises kVA dry-type transformer is presented in Tables 4 and 5,
in the coils and core of the transformer. respectively.
The results show that the middle leg of the transformer
4.4 Experimental tests and results core and coils are hotter than the other side of the trans-
former due to the electromagnetic effect. As shown in the
The transformer life is directly related to the insulation life above figures, the experimental results give approximate
of the transformers. The temperature rises and duration results with the simulation study.
causing the transformer’s insulation deterioration give
significant information about the lifetime of the trans-
former. In this section, the transformer coils and core 5 Results and discussion
temperature measurement tests of the manufactured dry-
type transformers of different powers are realized to verify The optimization methods are the key process to obtain the
the experimental results with the simulation results. J-type optimum value of the cost and efficiency. The transformers
thermocouples are used for measurement temperature at are optimized by using effective optimization methods.

123
Neural Computing and Applications (2022) 34:1079–1098 1093

Fig. 12 3 kVA 380/220 V delta-wye connected dry-type transformer a HV and LV voltages b HV and LV currents

123
1094 Neural Computing and Applications (2022) 34:1079–1098

Fig. 12 continued

Fig. 13 3 kVA 380/220 V delta-wye dry-type transformer losses in the transformer core

The optimization process is realized depending on the core the HS and PSO algorithms are more effective than pre-
quality factor and current density of the coils. During the vious studies in terms of cost and efficiency optimization.
optimization process, the highest efficiency and the lowest Transformers are modeled and analyzed in ANSYS/
cost values are obtained by different C and s values. Maxwell program by using optimum design parameters.
Therefore, the transformer manufacturers should decide the Thus, the electromagnetic and thermal analyses of the
s value by considering the economic criteria. transformers under full load conditions are realized. It is
In this study, the HS algorithm gives the best results in aimed to obtain realistic results by selecting the charac-
cost basis, and the PSO algorithm gives the best results in teristics of the materials in detail. In this way, the magnetic
efficiency basis obtained for 3 kVA and 5 kVA 3-phase flux flowing through the transformers, current–voltage
dry-type transformers. The optimization results show that signals of the transformers and loss analysis are obtained.
Transformers tested under full load conditions. Thus, the

123
Neural Computing and Applications (2022) 34:1079–1098 1095

Fig. 14 5 kVA 380/220 V Dyn 11 connected dry-type transformer losses in the core

Core Loss Model

Total Power
Resistance

Fig. 15 Thermal model of the transformer and thermal specifications of the core and coils

Fig. 16 The thermal analysis of 3 kVA dry-type transformer core and coils

123
1096 Neural Computing and Applications (2022) 34:1079–1098

Fig. 17 The temperature distribution of 5 kVA dry-type transformer core and coils

temperature distributions in the coils and cores of the

transformers were examined. Obtaining the electromag-
netic and thermal results via the program, testing of
transformers was enabled before the production of them.

6 Conclusion

Dry-type transformers have an air cooling system for their

coils and the core. Compared to the oil-filled transformers
with the same voltage/capacity rating, they are manufac-
tured by using larger coils. This case results in the over-
heating and the higher costs of the transformers due to
Fig. 18 The thermocouples connections and loaded operation test of
having natural cooling mechanisms. In this study, 3-phase
the manufactured transformer dry-type transformers with different voltage capacity rat-
ings are optimized in terms of cost and efficiency charac-
teristics by multiobjective swarm intelligence-based

Table 3 Cable connections

Channels Temperature Measurement Points
between the measurement
device and transformers Channel 1 The measurement device is connected with the transformer’s left leg and secondary side
CH-1
Channel 2 The measurement device is connected with the transformer’s left leg and primary side
CH-2
Channel 3 The measurement device is connected with the transformer’s right leg and secondary side
CH-3
Channel 4 The measurement device is connected with the transformer’s right leg and primary side
CH-4
Channel 5 The measurement device is connected with the transformer’s middle leg and secondary side
CH-5
Channel 6 The measurement device is connected with the transformer’s middle leg and primary side
CH-6
Channel 7 The measurement device is connected with the transformer’s middle leg and middle core
CH-7

123
Neural Computing and Applications (2022) 34:1079–1098 1097

Fig. 19 The temperature distribution-time graph of the transformers under the full load condition a 3 kVA b 5 kVA

optimization methods. The transformers are also analyzed

Table 4 The comparative of experimental and simulation temperature and tested electromagnetically and thermally. The results
measurement of 3 kVA dry-type transformer show that the core quality factor and current density
Channels The experimental The simulation Error capacity of the coils are the key values to optimize the
temperature temperature (%) parameters of the transformers. These values are directly
measurement (C) measurement (C)
linked to the efficiency and the losses related to tempera-
Channel 68.7 71.3 3.64 ture rises on the transformer coils and core. The opti-
1 mization methods achieve to obtain personal best and
Channel 65.5 70.0 6.42 global best values. The electromagnetic and thermal results
2 are verified with the experimental results in high accuracy.
Channel 65.9 70.9 7.05 This study plays a significant role in the transformer
3
manufacture process because it enables to obtain the opti-
Channel 64.8 69.9 7.29
mum cost, efficiency and thermal analysis before trans-
4
formers are manufactured.
Channel 67.8 71.4 5.04
5 In future studies, it is aimed to make a cost-size-effi-
Channel 66.8 70.7 5.51 ciency analysis of high power 3-phase dry-type trans-
6 formers. Also, it is aimed to examine the effect of the oil
Channel 69.8 72.0 3.05 cooling system on the temperature distribution in oil-filled
7 transformers.

Acknowledgements The author would like to acknowledge the Sci-

entific and Technological Research Council of Turkey (TUBITAK
Table 5 The comparative of experimental and simulation temperature
Project Number: 118E697) for full financial support.
measurement of 5 kVA dry-type transformer
Channels The experimental The simulation Error Funding The authors declare the following financial interests/perso-
temperature temperature (%) nal relationships which may be considered as potential competing
measurement (C) measurement (C) interests. This work was supported by the Scientific and Technolo-
gical Research Council of Turkey (TUBITAK) under Grant 118E697.
Channel 74.0 77.8 4.88
1 Declarations
Channel 72.1 77.4 6.84
2 Conflict of interests The authors declare that they have no known
Channel 75.7 78.8 3.93 competing financial interests or personal relationships that could have
3 appeared to influence the work reported in this paper.
Channel 73.6 78.2 5.88
4
Channel 78.4 79.9 1.87 References
5
Channel 77.5 79.7 2.76 1. Eslamian M, Vahidi B, Eslamian A (2011) Thermal analysis of
6 cast-resin dry-type transformers. Energy Convers Manage
Channel 79.4 80.3 1.12 52:2479–2488
7

123
1098 Neural Computing and Applications (2022) 34:1079–1098

2. Constantin D, Nicolae P-M, Nitu C-M (2013) 3D Finite element 16. Zile M (2019) Design of Power Transformer Core Using Created
analysis of a three phase power transformer. In: Eurocon 2013. Ant/Firefly Hybrid Optimization Algorithm. International Journal
IEEE, pp 1548–1552 on Technical and Physical Problems of Engineering (IJTPE)
3. Tapiawala G, Mishra RK (2016) Comprehensive modeling of dry 33–38
type foil winding transformer to analyse inter turn insulation 17. Hultmann Ayala HV, Dos Santos CL, Mariani VC, Luz VFD, M,
under lightning impulse voltage. In: 2016 National Power Sys- Leite JV, (2015) Harmony search approach based on ricker map
tems Conference (NPSC). IEEE, pp 1–5 for multi-objective transformer design optimization. IEEE Trans
4. Dawood K, Cinar MA, Alboyaci B, Sonmez O (2017) Modelling Magn 51:1–4. https://doi.org/10.1109/TMAG.2014.2372332
and analysis of transformer using numerical and analytical 18. Zile M (2019) Routine test analysis in power transformers by
methods. In: 2017 18th International Symposium on Electro- using firefly algorithm and computer program. IEEE Access
magnetic Fields in Mechatronics, Electrical and Electronic 7:132033–132040. https://doi.org/10.1109/ACCESS.2019.
Engineering (ISEF) Book of Abstracts. IEEE, pp 1–2 2939764
5. Răduca M, Haţiegan C, Pop N, Răduca E, Gillich G-R (2014) 19. Mehta HD, Patel RM (2014) A review on transformer design
Finite element analysis of heat transfer in transformers from high optimization and performance analysis using artificial intelli-
voltage stations. J Therm Anal Calorim 118:1355–1360 gence techniques. International Journal of Science and Research
6. Yan X, Li Y, Cheng Z, Zhang C (2020) Prediction of temperature 3:726–733
rise for ventilated dry-type transformer by 3-D coupled magneto- 20. Zile M (2020) Optimal design of power transformer tank using
fluid-thermal model. JAE 62:725–735. https://doi.org/10.3233/ ant/firefly hybrid heuristic algorithm. Turkish J Eng 4:17
JAE-190054 21. Aksu İÖ, Demirdelen T (2018) A comprehensive study on dry
7. Shaheen AM, El-Sehiemy RA, Farrag SM (2019) A reactive type transformer design with swarm-based metaheuristic opti-
power planning procedure considering iterative identification of mization methods for industrial applications. Energy Sourc Part
VAR candidate buses. Neural Comput Appl 31:653–674 A Recov Util Environ Effects 40:1743–1752
8. Reddy SS, Bijwe PR (2019) Differential evolution-based efficient 22. Demirdelen T (2019) Optimal design and experimental validation
multi-objective optimal power flow. Neural Comput Appl long-lasting, low loss transformer for low power renewable
31:509–522 energy system. Energy Sourc Part A Recov Util Environ Effects
9. Li P, Jiang J, Liu S, Fu P, Huang Y, Hui X, Qiang L, Hongyan Z 41:2534–2548
Electromagnetic-thermal analysis of 400 kV power transformer 23. Esenboga B, Demirdelen T (2020) Efficiency and cost based
on ITER PPEN substation. 10 multi-optimization and thermal/electromagnetic analyses of
10. Mikhak-Beyranvand M, Faiz J, Rezaei-Zare A, Rezaeealam B 3-Phase dry-type transformer. IETE J Res, 1–13
(2020) Electromagnetic and thermal behavior of a single-phase 24. Reddy JN, Gartling DK (2010) The finite element method in heat
transformer during Ferroresonance considering hysteresis model transfer and fluid dynamics. CRC Press
of core. Int J Electr Power Energy Syst 121:106078. https://doi. 25. Tsili MA, Amoiralis EI, Kladas AG, Souflaris AT (2012) Power
org/10.1016/j.ijepes.2020.106078 transformer thermal analysis by using an advanced coupled 3D
11. Orosz T, Pánek D, Karban P (2020) FEM Based Preliminary heat transfer and fluid flow FEM model. Int J Therm Sci
Design Optimization in Case of Large Power Transformers. Appl 53:188–201
Sci 10:1361. https://doi.org/10.3390/app10041361 26. Özüpak Y, MAMİŞ MS (2018) Realization of 3D electromag-
12. Vibhuti WG, Bhalla D (2020) Assessment of the use of FEM for netic and thermal analyzes of the autotransformer by finite ele-
computation of Electromagnetic Forces, Losses and Design of ments method. Int J Eng Comput Sci 7:24152–24157
Transformers. J Phys Conf Ser 1478:012029. https://doi.org/10. 27. Kennedy J, Eberhart R (1995) Particle swarm optimization, In:
1088/1742-6596/1478/1/012029 Proceedings of ICNN’95-International Conference on Neural
13. Tamilselvi S, Baskar S, Anandapadmanaban L, Karthikeyan V, Networks 4: 1942–1948
Rajasekar S (2018) Multi objective evolutionary algorithm for 28. Yang XS (2008) Firefly algorithm. Nat-inspir Metaheur Algo-
designing energy efficient distribution transformers. Swarm Evol rithms 20:79–90
Comput 42:109–124 29. Geem ZW, Kim JH, Loganathan GV (2001) A new heuristic
14. Azizian D, Bigdeli M (2014) Cast-resin dry-type transformer optimization algorithm: harmony search. SIMULATION
thermal modeling based on particle swarm optimization. In: 76:60–68
International Workshop Soft Computing Applications. Springer,
pp 141–153 Publisher’s Note Springer Nature remains neutral with regard to
15. Azizian D, Bigdeli M, Faiz J (2016) Design optimization of cast- jurisdictional claims in published maps and institutional affiliations.
resin transformer using nature-inspired algorithms. Arab J Sci
Eng 41:3491–3500

123