ISSN (Print) : 0974-6846

Indian Journal of Science and Technology, Vol 8(12), 59187, June 2015 ISSN (Online) : 0974-5645

Optimal Design of Transformer using Tournament

Selection based Elitist Genetic Algorithms
H. D. Mehta1,2* and Rajesh Patel3
School of Engineering, R.K. University, Rajkot - 360 020, Gujarat, India; [email protected]
1

2
Electrical Engineering Department, L.D. Engineering College, Ahmedabad - 380 015, Gujarat, India
3
HJD Institute of Technical Education and Research, Kera, Kachchh - 370 430, Gujarat, India;
[email protected]

Abstract
Transformers are vital components of power systems and its design requires reliable and rigorous solution methods. Optimal
design of transformer involves determination of design variables to optimize a particular objective, satisfying a set of con-
straints. This paper addresses the problem of optimal transformer design of a three phase core type distribution transformer
using Elitist Genetic Algorithms. Two MATLAB programs have been developed to accomplish the task. The first program im-
plements unconstrained minimization of the following four objective functions: total active part cost, total losses, percentage
impedance and transformer tank volume using GA; while second program considers both GA and conventional method to
minimize the active part cost while simultaneously satisfying BEE (Bureau of Energy Efficiency) standards and constraints.
After performing exhaustive analysis and comparing the results with those obtained by conventional method it is concluded
that the results obtained by Tournament Selection based Elitist Genetic Algorithm is near optimum. A design example on a
100 kVA, three phase core type distribution transformer using GA and conventional method is presented for illustration.

Keywords: Elitism, Genetic Algorithms, Optimal Transformer Design, Tournament Selection

1. Introduction by Jabr8 in which GP optimizer was used to design the

transformer operating at 100 kHz and at 60 Hz. However
Transformer design is a complex task which requires the as suggested in 9, it has two drawbacks (a) It requires large
knowledge of magnetic circuits, electromagnetism, electric of number of coefficients in polynomial approximations
circuit analysis, loss calculations and heat transfer. There and (b) Mathematical model is required to be developed
are more than 65 standards, 50 books and 400 published for each specific transformer type in advance.
articles in the domain of transformers1. The main aim of Transformer design consists of highly interrelated and
the design engineer is to optimize a particular objective heterogeneous design parameters10,11. A design is devel-
function depending upon the user requirement. In trans- oped after certain trials and errors and by experienced
former design optimization studies, much of the effort has judgment. Many design aids in the form of charts, curves,
been devoted to minimize the transformer manufactur- empirical constants and formulas have been created by
ing cost2–4 or active part cost5,6. Transformer design using experienced designers to minimize difficult calculations
multiple design method7 iteratively assigns different val- and to develop short cuts based on experience. However,
ues to transformer design variables, so as to generate large the transformer design procedure basically depends on
number of alternative designs. Finally the design which engineering’s judgment12,13.
satisfies all the constraints with the optimum value of Whatever the chosen design optimization method is,
objective function is selected; however this ­technique may the crux of the problem is to include how much detail in
fail to find the global optimum2. Transformer design opti- the problem description. Although, the main aim of design
mization using Geometric Programming was employed optimization is to find the lowest cost, the solution should

*Author for correspondence

 Optimal Design of Transformer using Tournament Selection based Elitist Genetic Algorithms

be such that the actual design can be produced with little 2. Transformer Design Procedure
additional work. Further, one should also concentrate
minimization of total losses, percentage impedance and This section describes a brief outline of the design method-
transformer tank dimensions as they are very critical to ology of a three phase core type distribution transformer
overall efficiency, voltage regulation and available space Some of the important assumptions made for the design
respectively. are mentioned below: Transformer LV and HV coils are
The studies carried out in 2–4 and 6–9 deal with ­optimization wound with aluminum conductors, as aluminum is found
of shell type transformer, and very less attention has been to be more economical than copper for transformers
devoted to optimal design of core type transformers. Design having rating of less than 190 kVA17. The core material
optimization using GA proposed in 5 does not give any idea is assumed to be of M4 grade, with a stacking factor of
regarding type of selection operator or type of crossover 0.97, and lamination thickness of 0.27 mm. Operating
mechanism adopted for optimization process. frequency is 50 Hz.
The main motive behind using GA for transformer
design optimization problem is due to the fact that GA’s 2.1 Calculation of Number of Turns for LV
have proved their mettle in solving various optimization and HV
problems such as unit commitment14, reactive power
In a transformer, voltage per turn is calculated using the
planning studies15, optimal DG placement in distribution
equation Et = K S , where Et is volt per turn and the
network16 etc.
value of K is constant given by 18
This paper proposes GA based design ­methodology
capable of minimizing (I) active part cost (II) total losses K = (4.44fΦm/AT × 103) 1/2 (1)
(III) percentage impedance and (IV) tank volume, using The number of turns in LV (NLV) and HV (NHV) are
three different selection operators for core type trans- then calculated as follows
formers. A MATLAB program has been developed
which allows the user to achieve any one of the objectives
­mentioned above.
N LV = VLV ( 3 × Et )
Further, if the designer selects the option of ­minimizing
the active part cost, another alternative is available to the N HV = 3 × VHV × N LV / VLV
user where a second MATLAB program using constrained
GA has been developed which minimizes the active part 2.2 Core Area and Diameter
cost, while simultaneously satisfying the BEE (Bureau of
The gross core area is calculated using the equation
Energy Efficiency) standards and constraints.
The major highlights of this paper are: Ag = (Et × 102) / (2.22 × Bmax × Kf) (2)
The value of Kf is assumed to be 0.97. The transformer
• For the first time comparative analysis of various
diameter assuming 9-stepped core is obtained from 19
­selection operators (i.e Roulette wheel selection,
Stochastic remainder roulette wheel selection and Ag × 4
Tournament selection) has been presented for dc =  (3)
p × 0.935
­transformer design optimization problem
• Elitism operator has been implemented, which ensures The core diameter obtained from equation (3) is then
that optimal value of an objective function once found rounded off to the nearest value
in any generation never worsens with successive
­generations 2.3 Calculation of Core Weight and Cost
• Statistical analysis for optimization problem has Transformer core weight can be obtained from 19
using
been carried out and the results obtained are com- the equation
pared with those obtained from MDM method to
ascertain the superiority of GA over conventional Wc = (4×Clc+3×Hw) ×Ag×Kf×ρc (4)
method for Transformer Design Optimization The core cost is then obtained by multiplying suitable
problem cost co-efficient with the core weight

2 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 H. D. Mehta and Rajesh Patel

2.4 Conductor Weight and Cost Iw = Wi / ( 3 × VLV )

The total conductor weight in a transformer depends
upon the mean diameter of LV and HV windings, total I0 = (Iμ2 + Iw2)½
number of turns, cross sectional area and the density of
winding material. It is given by 2.7 Percentage Reactance, Resistance and
Impedance Calculation
Wal = 3 ×ρal× π × (2 × MDLV× NLV × ALV + MDHV
× NHV × AHV) × 10–6 (5) Percentage reactance, resistance and impedance are
­calculated using the following relations
Once the winding weight is obtained, its cost can be
calculated by multiplying it with suitable cost co-efficient. 7.91 × f × Is × NLV2 × π × DM × R + R BLV
%X = (a + BHV )  (12)
The factor of ‘2’ appears in equation (5) as two strips of VL V × AsI × 106 3
LV are used.
% R = (LLLV + LLHV) ×100/S (13)

2.5 Load Losses of LV and HV Winding %Z = % X 2 + %R 2  (14)

The load losses of LV and HV winding, are calculated
using the following equations 2.8 Efficiency and Voltage Regulation
LLLV = 3×Is2× π× MDLV×NLV ×ρR/ ALV (6) The efficiency η at full load for power factor of cosΦ is
given by
LLHV = 3×Ip2× π× MDHV× NHV ×ρR/ AHV (7) η = (S× cosΦ)/( S× cosΦ + Wnl + Wfl) (15)

The total full load losses Wfl is then obtained by The percentage voltage regulation Vr at different
­values of power factors is given by
Wfl = LLLV + LLHV (8)
Vr = % R × cos φ + % X×sin φ (16)

2.6 No Load Loss and No Load Current

Calculation
3. Transformer Design
Optimization using
The core loss curve for M4 grade material which gives
specific no load loss Wnlsp at different values of flux density Conventional Method
is converted into fourth order equation using MATLAB This section describes the method for optimal design of a
polyfit function as demonstrated below three phase core type distribution transformer using Multiple
Wnlsp = 1.5291Bmax4 – 5.9664Bmax3 + 8.6933Bmax2 Design Methodology (MDM). This method is basically a
– 4.9237Bmax + 1.0388 (9) heuristic technique that assigns many ­alternative values to
the design variables so as to generate large ­number of alter-
The total no load loss, is then obtained by native designs13. Finally the design which satisfies all the
Wnl = Wnlsp ×Wc (10) problem constraints with minimum manufacturing cost of
active materials (cost of aluminium and CRGO) is selected.
Similarly, the curve of exciting volt-amperes versus This method optimizes the design of transformer with
flux density is converted into fourth order equation to the following technical characteristics
obtain exciting volt amperes we
• Three-phase oil immersed distribution transformers
We = 8.8542Bmax4 – 36.3249Bmax3 + 54.6091Bmax2 • Magnetic circuit of core type transformers
– 34.8050Bmax + 8.1222 (11) • Rectangular wire consisting of two strips for LV
­conductors and round cross sectional conductors for
The magnetizing component, core loss component
HV conductors.
and no load current are then obtained as shown below

Iµ = We × Wc / ( 3 × VLV ) The computer program takes into account many

­variations in design variables. These variations permit

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 3
 Optimal Design of Transformer using Tournament Selection based Elitist Genetic Algorithms

the investigation of a candidate solution. For each one • Value of current density in HV winding
of the candidate solutions, it is checked whether all • Value of current density in LV winding
constraints are satisfied, and if they are satisfied, the
cost of active materials is estimated and the solution Giving different values to ‘K’, ‘Bmax’, ‘Current density in LV
is considered as acceptable. Finally, among the accept- and HV’, the total candidate solutions (loops of the ­computer
able solutions, the transformer with the minimum program) are calculated from the following sequence
manufacturing cost is selected, which is the optimum Loops = Different values of K*No. of values of
transformer. In this design it has been assumed that Bmax*No. of values of current density
cost of aluminium is Rs. 177 per kg and the cost of For step size of 0.01, the program takes into account
CRGO (M4 grade, 0.27 mm thickness of ­lamination) is 16 different values of ‘K’, 51 different values of flux den-
Rs. 210 per kg. sity ‘Bmax’ and 5 different values of current density ‘δ’
There are four design variables that are taken into in LV and HV. Hence, total number of designs that are
account calculated by the program are 16×51×5×5 = 20400. The
flowchart for active part cost minimization is shown in
• Value of constant ‘K’ Figure 1, while results for cost minimization obtained
• Value of maximum flux density ‘Bmax’ from this method are depicted in Table 4.

START

Input data
kVA
HV, LV volts, frequency
Type of connection,
Percentage tappings
Cost of CRGO and Aluminium
Stacking factor
Enter max. limits on NLL,LL and
Percentage impedance

Calculate volts per turn

Calculate the dimensions of core
Calculate coil dimensions and its insulation
NO
Calculate winding and window height
Calculate yoke length and limb length
YES
Calculate transformer impedance K steps = Bm steps = į steps =
Calculate core weight and no load losses 0.32 to 1.2 to 1.7, 1.5 to
Calculate conductor weight 0.47, Step Step size = 1.54,Step
Calculate load losses size = 0.01 0.01 size = 0.01
Calculate total losses
Calculate efficiency
Calculate voltage regulation
Calculate cost of active materials
Calculate tank dimensions

NO
Does the design Transformer designs that do not satisfy the
meet constraints? given constraints are rejected

YES

Design accepted

Find the transformer design with

minimum cost

END

Figure 1. Flowchart for Multiple Design Methodology.

4 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 H. D. Mehta and Rajesh Patel

4. Genetic Algorithms Table 1. Performance of various selection operators

for TDO problem
Genetic Algorithms are based on random search method Sr. No Selection Objective Best Worst Mean Standard
that can be used to optimize complex problems. Some Operator Function value value Deviation
fundamental ideas of genetics are borrowed and used 1 RWS Active part 45662 47554 46516 499.82
cost (INR)
artificially to construct search algorithms that are robust Total 1622.39 1703.94 1657.67 20.50
and require minimal problem information. The main losses(Watts)
advantages of GA are Percentage 3.192 3.330 3.259 0.037
Impedance
• GA’s do not need a good initial estimation for the Tank Volume 228358 244905 231970 3978
(cm3)
sake of problem solution. In other words, if the ­initial 2 SRWS Active part 45545 46841 45977 327.18
estimates are weak, they can be corrected by an cost (INR)
­evolutionary process of fitness. Total losses 1616.23 1662.35 1638.26 16.50
• GA’s explore several areas of the search space (Watts)
Percentage 3.191 3.270 3.211 0.027
­simultaneously because of its population based Impedance
approach, which reduces the probability of being Tank Volume 227354 230958 226139 917.45
trapped in local optimum (cm3)
3 TS Active part 45165 45207 45174 12.08
• GA’s do not require any prior knowledge or ­properties
cost (INR)
of the function to be optimized such as convexity, Total losses 1612.37 1639.85 1614.34 6.040
smoothness, modality or existence of derivatives20 (Watts)
Percentage 3.176 3.181 3.178 0.002
The following three sub-sections describe basic operators Impedance
of Genetic Algorithms Tank Volume 226098 227210 226289 403.96
(cm3)

4.1 Reproduction in the literature of GAs. In most crossover operators, two

Reproduction is the first operator applied on population. individual strings (designs) are picked (or selected) at ran-
The reproduction operator is also called the selection dom from the mating pool generated by the ­reproduction
operator because it selects good strings of the population. operator and some portions of the strings are exchanged
The reproduction operator is used to pick above-aver- between the strings. A single point crossover preserves
age strings from the current population and insert their the structure of parent strings to the maximum extent
multiple copies in the mating pool based on probabilis- in child string. However, the preservation reduces with
tic procedure. Performance of three different selection the increase of cross sites and is minimum in case of uni-
operators for unconstrained optimization have been form crossover21. In this paper, crossover is done at four
compared as shown in Table 1 and the best ­performing ­different points along the chromosome length, which
selection operator is then utilized for Transformer combines the advantage of multipoint crossover and at
Design Optimization (TDO) as per BEE standards and the same time helps in preserving some portion of parent
constraints. Elitism has been employed in all the three strings.
selection operators. A copy of elite individual is not only
stored but it also takes part in crossover and mutation. If 4.3 Mutation
a better individual is found, it replaces the current elite The mutation operator is applied to the new strings with
individual; otherwise the same elite individual is carried a specific small mutation probability, pm. The need for
over to the next generation. mutation is to maintain diversity in the population. The
mutation operator changes the binary digit (allele’s value)
4.2 Crossover 1 to 0 and vice versa. In this paper, single point mutation
After reproduction, the crossover operator is has been used in which a mutation site is selected at ran-
­implemented. The purpose of crossover is to create new dom along the string length and the binary digit at that
strings by exchanging information among strings of the site is then changed from 1 to 0 or 0 to 1 with a probability
mating pool. Many crossover operators have been used of pm.

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 5
 Optimal Design of Transformer using Tournament Selection based Elitist Genetic Algorithms

5. Implementation of GA Table 2. Input parameters for 1-star and 2-star rated

transformers
Technique for Transformer
Design Sr.No Parameter 1-star 2-star Units

This section describes the methods for optimal design 1 Rated power 100 100 kVA
of 100 kVA, 11/0.433 kV, distribution transformer using 2 Max. Total Losses 2020 1910 W
Genetic Algorithms. The main advantage of GA is that permitted
different objective functions can be optimized with little
modification in the program. Two MATLAB programs 3 Max. Losses permitted 700 610 W
at half load
have been developed to avoid complexity and to maintain
clarity. 4 Max. NLL permitted 220 200 W

Method-I 5 Percentage impedance 4.7 4.7 %

The first MATLAB program implements unconstrained permitted
GA technique to minimize any one of the four objectives
6 Rated low voltage 433 433 V
namely (1) Active part cost (2) Total losses (3) Percentage
impedance (4) Transformer tank volume. The user can 7 Rated high voltage 11000 11000 V
select any one of the above mentioned objective as per
requirement. 8 Temperature rise 50 50 0
C

Method-II
tive functions using unconstrained GA ­optimization
The second MATLAB program considers the ­constraints
technique. The choice is left to the user to decide any one
pertaining to IS 2026 and IS 1180 (part 1) and ­implements
objective as per requirement. The second part minimizes
constrained GA technique to minimize active part
the active part cost, subject to constraints specified in 22
cost of a transformer. This transformer design satisfies
for 1-star and 2-star rated transformers. Although 22 does
1-star and 2-star rating of distribution transformer as
specify general guidelines about the limiting values of
per BEE (Bureau of Energy Efficiency) standards and
total losses at full load and half load for different star rat-
­specifications22.
ing of transformers, information regarding bifurcation of
The design inputs for design of 100 kVA, 11/0.433
no-load and load losses is not available. Therefore, their
kV, Dyn-11 distribution transformer are shown in
limiting values mentioned in Table 2 are selected as per
Table 2. The control parameters for GA are: Population
customer’s requirement.
size = 40, Max. Generations = 100, Crossover probability
= 0.8, Mutation probability = 0.02, No. of dimensions =
4, Chromosome length = 80, Elite count = 1. It should 6.1 Minimization of Various Objectives
be noted that input parameters mentioned in Table-1 are using Unconstrained Genetic
used as inputs to minimize active part cost as per BEE Algorithms
standards and constraints. For unconstrained transformer Table 1 shows the performance of various selection
design ­optimization, no limits are imposed on total losses, ­operators for unconstrained TDO problem. By trial and
no-load losses, and percentage impedance. error, it was found that 20 trial runs were sufficient for
After trial and error, it was found that population size assessing the performance analysis of selection opera-
of 40 and a number of 100 generations, with crossover tors. As evident from Table 1, Tournament Selection
probability of 0.8 and mutation probability of 0.02 ­provide is the most reliable selection operator in terms of best
good results for TDO. value, mean and standard deviation. Figure 2 to Figure 5
indicates the optimum value of each objective function,
obtained in each generation using TS operator. Presence
6. Results and Discussion of elitism ensures that the optimum value of objective
This section has been divided into two parts. The first part function, once obtained in particular generation is not
demonstrates the results for minimizing different objec- lost in s­ uccessive generations.

6 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 H. D. Mehta and Rajesh Patel

5
x 10
4
A c ti ve pa rt c o s t fo u nd i n e a c h ge ne ra ti o n
6.2 Active Part Cost Minimization
using MDM and Constrained
4. 95

4. 9

Genetic Algorithms
4. 85

4. 8
Active material cost

4. 75

4. 7

4. 65 Since cost minimization is the prime requirement in any

optimization process, this section deals with minimization
4. 6

4. 55

4. 5
of active part cost of a transformer, while simultaneously
4. 45
0 10 20 30 40 50 60
Number of generations
70 80 90 100
satisfying BEE standards and constraints for 1-star and
2-star rated transformers. Table 2 demonstrates the inputs
Figure 2. Variation of Active Part Cost (INR) with
to the program while Table 4 exemplifies important design
generations using TS operator.
dimensions and performance parameters of 100 kVA,
11/0.433 kV Dyn-11 transformer obtained by MDM and
1700
T o ta l l o s s e s fo u nd i n e a c h ge ne ra ti o n
TS based GA method. The value of penalty factors must
1690

1680
be chosen judiciously and it requires extensive experi-
1670 mentation21. A very high value of penalty factor quickly
1660
helps in steering GA towards convergence, however
Total losses

1650

1640 sometimes it may converge to a local optimum because

1630

1620
of high selection pressure. On the contrary, a low value
1610 of penalty factor helps more effectively in ­exploring the
1600
0 10 20 30 40 50 60
Number of generations
70 80 90 100
search space; however it may sometimes lead to ­infeasible
solution because of low selection pressure21.
Figure 3. Variation of Total losses (watts) with generations In this paper, initially all penalty factors are set to zero.
using TS operator. If an infeasible solution is generated, penalty is imposed
as shown in Table 3. After performing number of trials,
P e rc e nta ge i m pe da nc e fo u nd i n e a c h ge ne ra ti o n the value of penalty factors tuned for different objective
3. 5

3. 45 functions are depicted in Table 3. From Table 4 it is evi-

3. 4

3. 35
dent that GA is able to obtain solution which is 2.71%
and 1.34% cheaper for 1-star and 2-star rated trans-
Percentage impedance

3. 3

3. 25

3. 2
formers ­respectively as compared to Multiple Design
3. 15 Methodology.
3. 1

Table 3. Penalty factors and objective function for

3. 05

3
0 10 20 30 40 50 60 70 80 90 100
Number of generations
constrained TDO
Figure 4. Variation of Percentage Impedance with Sr. No Penalty factor 1-star rated 2-star rated
generations using TS operator. transformer transformer

1 P_NLL (for Wnl > 1000*(Wnl - 1000*(Wnl -

2. 5
x 10
5
T a nk vo l u m e fo u nd i n e a c h ge ne ra ti o n NLL_max) NLL_max) NLL_max)
2. 45
2 P_TLL (for Wtl > 100*(Wtl – 100*(Wtl – TLL_
2. 4 TLL_max) TLL_max) max)
Tank volume

3 P_HLL(for Wthl > 100*(Wthl – 1000*(Wthl –

2. 35

2. 3
HLL_max) HLL_max) HLL_max)
2. 25

4 P_IM (for %Z > 40000*(%Z – 25000*(%Z –

PIM_max) PIM_max) PIM_max)
2. 2
0 10 20 30 40 50 60 70 80 90 100
Number of generations

Figure 5. Variation of Tank Volume (cm3) with generations Objective function F(x) = Total Active Part Cost + P_NLL +
using TS operator. P_TLL + P_HLL + P_IM

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 7
 Optimal Design of Transformer using Tournament Selection based Elitist Genetic Algorithms

Table 4. Main design dimensions and important 8. Acknowledgement

technical parameters of 1-star and 2-star rated
transformers by MDM and TS The authors would like to thank Mr. Vasant C. Vyas
Parameter Design of Design of Design of Design of Units (Proprietor, Jyoti Transelect, Co. Bhuj, India) Mr. P.H.
1-star rated 1-star rated 2-star rated 2-star rated Trivedi (Design Engineer, Jyoti Transelect. Co, Bhuj,
transformer transformer transformer transformer India) for support and helpful discussions.
by MDM by TS by MDM by TS
Active part cost 50051 48693 57921 57144 INR
No load losses 219.22 219.86 196.83 199.93 W 9. References
Load losses 1795.85 1799.12 1651.04 1639.97 W
Total losses 2015.07 2018.98 1847.87 1839.90 W 1. Amoiralis EI, Tsili MA, Kladas AG. Transformer design
Total half load 668.18 669.64 609.59 609.92 W and optimization: A Literature Survey. IEEE Transactions
losses on Power Delivery. 2009; 24(4):1999–24.
Volt per turn 3.400 3.268 3.600 3.649 ------ 2. Georgilakis PS, Olivares JC, Esparza-Gonzalez MS. An
HV turns 3419 3511 3188 3142 ------
Evolutionary computation solution to transformer design
LV turns 74 76 69 68 ------
Gross core area 100.56 95.41 116.90 117.20 cm2
optimization problem. 7th International Conference
Core limb centre 259 255 276 276 mm on Electrical and Electronics Engineering Research;
Total yoke length 1036 1020 1104 1104 mm Aguascalientes, Mexico. 2010. p. 226–31.
Total limb length 1470 1518 1410 1374 mm 3. Amoiralis EI, Georgilakis PS, Tsili MA, Kladas AG. Global
Core weight 187.01 179.69 218.09 215.51 kg transformer optimization method using evolutionary
Conductor 60.89 61.90 68.48 67.15 kg design and numerical field computation. IEEE Transactions
weight
on Magnetics. 2009; 45(3):1720–3.
Percentage 4.344 4.367 4.437 4.436 %
4. Georgilakis PS, Tsili MA, Souflaris AT. A heuristic solution
impedance
to transformer manufacturing cost optimization ­problem.
Tank length 81.4 80.2 86.5 86.5 cm
Tank breadth 32.6 32.2 34.3 34.3 cm Journals of Materials Processing Technology. 2007;
Tank height 90.2 91.3 90.0 88.8 cm 81(1-3):260–6.
Tank volume 239358 235776 267025 263465 cm3 5. Khatri A, Malik H, Rahi OP. Optimal Design of Power
Efficiency (full 98.02 98.02 98.18 98.19 % Transformer using Genetic Algorithm. Proceedings of
load, upf) International Conference on Communication Systems and
Network Technologies; 2012. p. 830–3.
6. Amoiralis EI, Tsili MA, Georgilakis PS, Kladas AG. A par-
allel mixed integer programming-finite element technique
7. Conclusion for global design optimization of power transformer. IEEE
transactions on Magnetics. 2008; 44(6):1022–5.
In this paper, design optimization of transformer using 7. Olivares JC, Georgilakis PS, Escarela-Perez R, Campero-
Genetic Algorithms and conventional method has been Littlewood E. Optimal design of single-phase shell type
demonstrated. Although, no constraints were imposed in distribution transformers based on a multiple design
first method (i.e. minimization of cost, total losses, per- method validated by measurements. Electr Engg. 2011;
centage impedance and tank dimensions), the program 93:237–46.
can be modified to accommodate any constraints, desired 8. Jabr RA. Application of Geometric Programming to
Transformer Design. IEEE Transactions on Magnetics.
by the user. The proposed method is very effective as GA’s
2005; 41(11):4261–9.
are more likely find the global optimum because of their
9. Georgilakis PS. Recursive genetic algorithm-finite ele-
population based approach. A saving of 2.71% and 1.34%
ment method technique for the solution of transformer
obtained by TS based GA method as compared to con- cost ­minimization problem. IET Electr Power Appl. 2009;
ventional method may not sound great, but considering 3(6):514–9.
the fact that the numbers of distribution transformers in 10. Grady WM, Chan R, Sarnotyj MJ, Ferraro RJ, Bierschenk
any region far exceed the number of power transformers JL. A PC-based computer program for teaching the design
in the same region, the cost benefits obtained from GA and analysis of dry type transformers. IEEE Transactions
based transformer design can be appreciated. Small trans- on Power Systems. 1992; 7(2):709–17.
former manufacturing companies and even ­inexperienced 11. Geromel LH, Souza CR. The application of intelligent
engineers can successfully use this software. systems in power transformer design. Proceedings of

8 Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology
 H. D. Mehta and Rajesh Patel

IEEE Canadian Conference Electrical and Computer genetic algorithm based method. Indian Journal of Science
Engineering; Winnipeg, Canada. 2002. p. 285–90. and Technology. 2012; 5(3):2281–6.
12. Heathcote MJ. The J & P Transformer Book. 12th ed. 17. Olivares JC, de Leon F, Georgilakis PS, Escarela-Perez
London, U.K: Newness; 1998. R. Selection of copper against aluminium windings for
13. Georgilakis PS. Spotlight on Modern Transformer Design. ­distribution transformers. IET Electr Power Appl. 2010;
London, U.K: Springer; 2009. 4(6):74–85.
14. Rouhi F, Effetnejad R. Unit Commitment in Power 18. Sawhney AK. A course in Electrical Machine Design. 6th
System by Combination of Dynamic Programming (DP), ed. Dhanpat Rai Publication & Co; 2006.
Genetic Algorithm (GA) and Particle Swarm Optimization 19. Dasgupta I. Design of Transformers. McGraw Hill Education
(PSO). Indian Journal of Science and Technology. 2015; (India) Private Limited; 2013.
8(2):134–41. 20. Padhy NP. Artificial Intelligence and Intelligent Systems.
15. Ambika R, Rajeshwari R, Nivedita A. Comparative Analysis Oxford University Press; 2005.
of Nature Inspired Algorithms Applied to Reactive Power 21. Deb K. Multiobjective Optimization using Evolutionary
Planning Studies. Indian Journal of Science and Technology. Algorithms. John Wiley & Sons; 2001.
2015; 8(5):445–53. 22. Notification of the Gazette of India. Part II, Sec 3 (ii), S.O.
16. Ali A, Mohsen D, Majid D. Optimal placement and Published. 12/1/2009; 185(E):44–9.
­estimation of DG capacity in distribution networks using

Appendix
List of symbols Wfl Total losses in LV and HV winding (watts)
Et Volt per turn Wnlsp Specific no-load losses (watts/kg)
S Rating of transformer (kVA) Wnl No load losses (watts)
f Frequency (Hz) Wtl Total losses (watts)
AT Ampere turns Wthl Total losses at half load (watts)
VLV Rating of LV winding (Volts) We Exciting volt-amperes (VA/kg)
VHV Rating of HV winding (Volts) Iμ Magnetizing component of current (Amp)
NLV Number of turns in LV winding Iw Core loss component of current (Amp)
NHV Number of turns in HV winding Dm Mean diameter of LV and HV coil (mm)
Ag Gross core area (cm2)
RBHV Radial build of HV winding (mm)
Bmax Maximum flux density in core (Wb/m2)
RBLV Radial build of LV winding (mm)
Kf Stacking factor
Asl Axial stack of LV and HV winding (mm)
dc Core diameter (cm)
Wc Core weight (kg) a Gap between LV and HV winding (mm)
Clc Core limb centre (cm) %X Percentage reactance
Hw Window height (cm) %R Percentage resistance
ρc Density of core material (gm/cm3) %Z Percentage impedance
Wal Weight of aluminium (kg) η Efficiency
ρal Density of aluminium (gm/cm3) RWS Roulette Wheel Selection
MDLV Mean diameter of LV winding (mm) SRWSS to chastic Remainder Roulette Wheel Selection
MDHV Mean diameter of HV winding (mm) TS Tournament Selection
ALV Cross sectional area of LV winding (mm2) NLL_max Maximum permitted no-load losses (watts)
AHV Cross sectional area of HV winding (mm2) TLL_max Maximum permitted total losses at full load
LLLV Load losses in LV winding (watts) (watts)
LLHV Load losses in HV winding (watts) HLL_max Maximum losses per permitted at half load
Is Rated current of LV winding (Amp) (watts)
Ip Rated current of HV winding (Amp) PIM_max Maximum allowed percentage i­ mpedance
ρR Resistivity of aluminium (ohm-cm) MDM Multiple Design Methodology

Vol 8 (12) | June 2015 | www.indjst.org Indian Journal of Science and Technology 9